diff --git a/3rdparty/spirv-tools/include/generated/DebugInfo.h b/3rdparty/spirv-tools/include/generated/DebugInfo.h index 76c61714c..a37e3a947 100644 --- a/3rdparty/spirv-tools/include/generated/DebugInfo.h +++ b/3rdparty/spirv-tools/include/generated/DebugInfo.h @@ -1,4 +1,4 @@ -// Copyright (c) 2017 The Khronos Group Inc. +// Copyright (c) 2017-2024 The Khronos Group Inc. // // Permission is hereby granted, free of charge, to any person obtaining a copy // of this software and/or associated documentation files (the "Materials"), diff --git a/3rdparty/spirv-tools/include/generated/NonSemanticShaderDebugInfo100.h b/3rdparty/spirv-tools/include/generated/NonSemanticShaderDebugInfo100.h index 9db6951a8..e09368e24 100644 --- a/3rdparty/spirv-tools/include/generated/NonSemanticShaderDebugInfo100.h +++ b/3rdparty/spirv-tools/include/generated/NonSemanticShaderDebugInfo100.h @@ -1,4 +1,4 @@ -// Copyright (c) 2018 The Khronos Group Inc. +// Copyright (c) 2018-2024 The Khronos Group Inc. // // Permission is hereby granted, free of charge, to any person obtaining a copy // of this software and/or associated documentation files (the "Materials"), diff --git a/3rdparty/spirv-tools/include/generated/OpenCLDebugInfo100.h b/3rdparty/spirv-tools/include/generated/OpenCLDebugInfo100.h index dcb06ed49..1da00991a 100644 --- a/3rdparty/spirv-tools/include/generated/OpenCLDebugInfo100.h +++ b/3rdparty/spirv-tools/include/generated/OpenCLDebugInfo100.h @@ -1,4 +1,4 @@ -// Copyright (c) 2018 The Khronos Group Inc. +// Copyright (c) 2018-2024 The Khronos Group Inc. // // Permission is hereby granted, free of charge, to any person obtaining a copy // of this software and/or associated documentation files (the "Materials"), diff --git a/3rdparty/spirv-tools/include/generated/build-version.inc b/3rdparty/spirv-tools/include/generated/build-version.inc index ff13767b1..4535672f8 100644 --- a/3rdparty/spirv-tools/include/generated/build-version.inc +++ b/3rdparty/spirv-tools/include/generated/build-version.inc @@ -1 +1 @@ -"v2023.6", "SPIRV-Tools v2023.6 v2023.6.rc1-7-g57aba7ff" +"v2023.6", "SPIRV-Tools v2023.6 v2023.6.rc1-37-g58b2f0b6" diff --git a/3rdparty/spirv-tools/include/generated/core.insts-unified1.inc b/3rdparty/spirv-tools/include/generated/core.insts-unified1.inc index e0ddf501a..de8d1e877 100644 --- a/3rdparty/spirv-tools/include/generated/core.insts-unified1.inc +++ b/3rdparty/spirv-tools/include/generated/core.insts-unified1.inc @@ -44,6 +44,7 @@ static const spv::Capability pygen_variable_caps_Kernel[] = {spv::Capability::Ke static const spv::Capability pygen_variable_caps_KernelImageQuery[] = {spv::Capability::Kernel, spv::Capability::ImageQuery}; static const spv::Capability pygen_variable_caps_LiteralSampler[] = {spv::Capability::LiteralSampler}; static const spv::Capability pygen_variable_caps_LongCompositesINTEL[] = {spv::Capability::LongCompositesINTEL}; +static const spv::Capability pygen_variable_caps_MaskedGatherScatterINTEL[] = {spv::Capability::MaskedGatherScatterINTEL}; static const spv::Capability pygen_variable_caps_Matrix[] = {spv::Capability::Matrix}; static const spv::Capability pygen_variable_caps_MemoryAccessAliasingINTEL[] = {spv::Capability::MemoryAccessAliasingINTEL}; static const spv::Capability pygen_variable_caps_MeshShadingEXT[] = {spv::Capability::MeshShadingEXT}; @@ -51,6 +52,7 @@ static const spv::Capability pygen_variable_caps_MeshShadingNV[] = {spv::Capabil static const spv::Capability pygen_variable_caps_NamedBarrier[] = {spv::Capability::NamedBarrier}; static const spv::Capability pygen_variable_caps_PipeStorage[] = {spv::Capability::PipeStorage}; static const spv::Capability pygen_variable_caps_Pipes[] = {spv::Capability::Pipes}; +static const spv::Capability pygen_variable_caps_QuadControlKHR[] = {spv::Capability::QuadControlKHR}; static const spv::Capability pygen_variable_caps_RayQueryKHR[] = {spv::Capability::RayQueryKHR}; static const spv::Capability pygen_variable_caps_RayQueryPositionFetchKHR[] = {spv::Capability::RayQueryPositionFetchKHR}; static const spv::Capability pygen_variable_caps_RayTracingKHR[] = {spv::Capability::RayTracingKHR}; @@ -520,6 +522,8 @@ static const spv_opcode_desc_t kOpcodeTableEntries[] = { {"FinalizeNodePayloadsAMDX", spv::Op::OpFinalizeNodePayloadsAMDX, 1, pygen_variable_caps_ShaderEnqueueAMDX, 1, {SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"FinishWritingNodePayloadAMDX", spv::Op::OpFinishWritingNodePayloadAMDX, 1, pygen_variable_caps_ShaderEnqueueAMDX, 3, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"InitializeNodePayloadsAMDX", spv::Op::OpInitializeNodePayloadsAMDX, 1, pygen_variable_caps_ShaderEnqueueAMDX, 4, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu}, + {"GroupNonUniformQuadAllKHR", spv::Op::OpGroupNonUniformQuadAllKHR, 1, pygen_variable_caps_QuadControlKHR, 3, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, + {"GroupNonUniformQuadAnyKHR", spv::Op::OpGroupNonUniformQuadAnyKHR, 1, pygen_variable_caps_QuadControlKHR, 3, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"HitObjectRecordHitMotionNV", spv::Op::OpHitObjectRecordHitMotionNV, 2, pygen_variable_caps_ShaderInvocationReorderNVRayTracingMotionBlurNV, 14, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"HitObjectRecordHitWithIndexMotionNV", spv::Op::OpHitObjectRecordHitWithIndexMotionNV, 2, pygen_variable_caps_ShaderInvocationReorderNVRayTracingMotionBlurNV, 13, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"HitObjectRecordMissMotionNV", spv::Op::OpHitObjectRecordMissMotionNV, 2, pygen_variable_caps_ShaderInvocationReorderNVRayTracingMotionBlurNV, 7, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu}, @@ -841,5 +845,7 @@ static const spv_opcode_desc_t kOpcodeTableEntries[] = { {"GroupBitwiseXorKHR", spv::Op::OpGroupBitwiseXorKHR, 1, pygen_variable_caps_GroupUniformArithmeticKHR, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_GROUP_OPERATION, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"GroupLogicalAndKHR", spv::Op::OpGroupLogicalAndKHR, 1, pygen_variable_caps_GroupUniformArithmeticKHR, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_GROUP_OPERATION, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, {"GroupLogicalOrKHR", spv::Op::OpGroupLogicalOrKHR, 1, pygen_variable_caps_GroupUniformArithmeticKHR, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_GROUP_OPERATION, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, - {"GroupLogicalXorKHR", spv::Op::OpGroupLogicalXorKHR, 1, pygen_variable_caps_GroupUniformArithmeticKHR, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_GROUP_OPERATION, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu} + {"GroupLogicalXorKHR", spv::Op::OpGroupLogicalXorKHR, 1, pygen_variable_caps_GroupUniformArithmeticKHR, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_GROUP_OPERATION, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, + {"MaskedGatherINTEL", spv::Op::OpMaskedGatherINTEL, 1, pygen_variable_caps_MaskedGatherScatterINTEL, 6, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_LITERAL_INTEGER, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu}, + {"MaskedScatterINTEL", spv::Op::OpMaskedScatterINTEL, 1, pygen_variable_caps_MaskedGatherScatterINTEL, 4, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_LITERAL_INTEGER, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu} }; \ No newline at end of file diff --git a/3rdparty/spirv-tools/include/generated/enum_string_mapping.inc b/3rdparty/spirv-tools/include/generated/enum_string_mapping.inc index 058853154..78e648283 100644 --- a/3rdparty/spirv-tools/include/generated/enum_string_mapping.inc +++ b/3rdparty/spirv-tools/include/generated/enum_string_mapping.inc @@ -120,6 +120,8 @@ const char* ExtensionToString(Extension extension) { return "SPV_INTEL_long_composites"; case Extension::kSPV_INTEL_loop_fuse: return "SPV_INTEL_loop_fuse"; + case Extension::kSPV_INTEL_masked_gather_scatter: + return "SPV_INTEL_masked_gather_scatter"; case Extension::kSPV_INTEL_media_block_io: return "SPV_INTEL_media_block_io"; case Extension::kSPV_INTEL_memory_access_aliasing: @@ -156,6 +158,8 @@ const char* ExtensionToString(Extension extension) { return "SPV_KHR_expect_assume"; case Extension::kSPV_KHR_float_controls: return "SPV_KHR_float_controls"; + case Extension::kSPV_KHR_float_controls2: + return "SPV_KHR_float_controls2"; case Extension::kSPV_KHR_fragment_shader_barycentric: return "SPV_KHR_fragment_shader_barycentric"; case Extension::kSPV_KHR_fragment_shading_rate: @@ -164,6 +168,8 @@ const char* ExtensionToString(Extension extension) { return "SPV_KHR_integer_dot_product"; case Extension::kSPV_KHR_linkonce_odr: return "SPV_KHR_linkonce_odr"; + case Extension::kSPV_KHR_maximal_reconvergence: + return "SPV_KHR_maximal_reconvergence"; case Extension::kSPV_KHR_multiview: return "SPV_KHR_multiview"; case Extension::kSPV_KHR_no_integer_wrap_decoration: @@ -174,6 +180,8 @@ const char* ExtensionToString(Extension extension) { return "SPV_KHR_physical_storage_buffer"; case Extension::kSPV_KHR_post_depth_coverage: return "SPV_KHR_post_depth_coverage"; + case Extension::kSPV_KHR_quad_control: + return "SPV_KHR_quad_control"; case Extension::kSPV_KHR_ray_cull_mask: return "SPV_KHR_ray_cull_mask"; case Extension::kSPV_KHR_ray_query: @@ -255,8 +263,8 @@ const char* ExtensionToString(Extension extension) { bool GetExtensionFromString(const char* str, Extension* extension) { - static const char* known_ext_strs[] = { "SPV_AMDX_shader_enqueue", "SPV_AMD_gcn_shader", "SPV_AMD_gpu_shader_half_float", "SPV_AMD_gpu_shader_half_float_fetch", "SPV_AMD_gpu_shader_int16", "SPV_AMD_shader_ballot", "SPV_AMD_shader_early_and_late_fragment_tests", "SPV_AMD_shader_explicit_vertex_parameter", "SPV_AMD_shader_fragment_mask", "SPV_AMD_shader_image_load_store_lod", "SPV_AMD_shader_trinary_minmax", "SPV_AMD_texture_gather_bias_lod", "SPV_ARM_core_builtins", "SPV_EXT_demote_to_helper_invocation", "SPV_EXT_descriptor_indexing", "SPV_EXT_fragment_fully_covered", "SPV_EXT_fragment_invocation_density", "SPV_EXT_fragment_shader_interlock", "SPV_EXT_mesh_shader", "SPV_EXT_opacity_micromap", "SPV_EXT_physical_storage_buffer", "SPV_EXT_shader_atomic_float16_add", "SPV_EXT_shader_atomic_float_add", "SPV_EXT_shader_atomic_float_min_max", "SPV_EXT_shader_image_int64", "SPV_EXT_shader_stencil_export", "SPV_EXT_shader_tile_image", "SPV_EXT_shader_viewport_index_layer", "SPV_GOOGLE_decorate_string", "SPV_GOOGLE_hlsl_functionality1", "SPV_GOOGLE_user_type", "SPV_INTEL_arbitrary_precision_fixed_point", "SPV_INTEL_arbitrary_precision_floating_point", "SPV_INTEL_arbitrary_precision_integers", "SPV_INTEL_bfloat16_conversion", "SPV_INTEL_blocking_pipes", "SPV_INTEL_cache_controls", "SPV_INTEL_debug_module", "SPV_INTEL_device_side_avc_motion_estimation", "SPV_INTEL_float_controls2", "SPV_INTEL_fp_fast_math_mode", "SPV_INTEL_fp_max_error", "SPV_INTEL_fpga_argument_interfaces", "SPV_INTEL_fpga_buffer_location", "SPV_INTEL_fpga_cluster_attributes", "SPV_INTEL_fpga_dsp_control", "SPV_INTEL_fpga_invocation_pipelining_attributes", "SPV_INTEL_fpga_latency_control", "SPV_INTEL_fpga_loop_controls", "SPV_INTEL_fpga_memory_accesses", "SPV_INTEL_fpga_memory_attributes", "SPV_INTEL_fpga_reg", "SPV_INTEL_function_pointers", "SPV_INTEL_global_variable_fpga_decorations", "SPV_INTEL_global_variable_host_access", "SPV_INTEL_inline_assembly", "SPV_INTEL_io_pipes", "SPV_INTEL_kernel_attributes", "SPV_INTEL_long_composites", "SPV_INTEL_loop_fuse", "SPV_INTEL_media_block_io", "SPV_INTEL_memory_access_aliasing", "SPV_INTEL_optnone", "SPV_INTEL_runtime_aligned", "SPV_INTEL_shader_integer_functions2", "SPV_INTEL_split_barrier", "SPV_INTEL_subgroups", "SPV_INTEL_unstructured_loop_controls", "SPV_INTEL_usm_storage_classes", "SPV_INTEL_variable_length_array", "SPV_INTEL_vector_compute", "SPV_KHR_16bit_storage", "SPV_KHR_8bit_storage", "SPV_KHR_bit_instructions", "SPV_KHR_cooperative_matrix", "SPV_KHR_device_group", "SPV_KHR_expect_assume", "SPV_KHR_float_controls", "SPV_KHR_fragment_shader_barycentric", "SPV_KHR_fragment_shading_rate", "SPV_KHR_integer_dot_product", "SPV_KHR_linkonce_odr", "SPV_KHR_multiview", "SPV_KHR_no_integer_wrap_decoration", "SPV_KHR_non_semantic_info", "SPV_KHR_physical_storage_buffer", "SPV_KHR_post_depth_coverage", "SPV_KHR_ray_cull_mask", "SPV_KHR_ray_query", "SPV_KHR_ray_tracing", "SPV_KHR_ray_tracing_position_fetch", "SPV_KHR_shader_atomic_counter_ops", "SPV_KHR_shader_ballot", "SPV_KHR_shader_clock", "SPV_KHR_shader_draw_parameters", "SPV_KHR_storage_buffer_storage_class", "SPV_KHR_subgroup_rotate", "SPV_KHR_subgroup_uniform_control_flow", "SPV_KHR_subgroup_vote", "SPV_KHR_terminate_invocation", "SPV_KHR_uniform_group_instructions", "SPV_KHR_variable_pointers", "SPV_KHR_vulkan_memory_model", "SPV_KHR_workgroup_memory_explicit_layout", "SPV_NVX_multiview_per_view_attributes", "SPV_NV_bindless_texture", "SPV_NV_compute_shader_derivatives", "SPV_NV_cooperative_matrix", "SPV_NV_displacement_micromap", "SPV_NV_fragment_shader_barycentric", "SPV_NV_geometry_shader_passthrough", "SPV_NV_mesh_shader", "SPV_NV_ray_tracing", "SPV_NV_ray_tracing_motion_blur", "SPV_NV_sample_mask_override_coverage", "SPV_NV_shader_image_footprint", "SPV_NV_shader_invocation_reorder", "SPV_NV_shader_sm_builtins", "SPV_NV_shader_subgroup_partitioned", "SPV_NV_shading_rate", "SPV_NV_stereo_view_rendering", "SPV_NV_viewport_array2", "SPV_QCOM_image_processing", "SPV_VALIDATOR_ignore_type_decl_unique" }; - static const Extension known_ext_ids[] = { Extension::kSPV_AMDX_shader_enqueue, Extension::kSPV_AMD_gcn_shader, Extension::kSPV_AMD_gpu_shader_half_float, Extension::kSPV_AMD_gpu_shader_half_float_fetch, Extension::kSPV_AMD_gpu_shader_int16, Extension::kSPV_AMD_shader_ballot, Extension::kSPV_AMD_shader_early_and_late_fragment_tests, Extension::kSPV_AMD_shader_explicit_vertex_parameter, Extension::kSPV_AMD_shader_fragment_mask, Extension::kSPV_AMD_shader_image_load_store_lod, Extension::kSPV_AMD_shader_trinary_minmax, Extension::kSPV_AMD_texture_gather_bias_lod, Extension::kSPV_ARM_core_builtins, Extension::kSPV_EXT_demote_to_helper_invocation, Extension::kSPV_EXT_descriptor_indexing, Extension::kSPV_EXT_fragment_fully_covered, Extension::kSPV_EXT_fragment_invocation_density, Extension::kSPV_EXT_fragment_shader_interlock, Extension::kSPV_EXT_mesh_shader, Extension::kSPV_EXT_opacity_micromap, Extension::kSPV_EXT_physical_storage_buffer, Extension::kSPV_EXT_shader_atomic_float16_add, Extension::kSPV_EXT_shader_atomic_float_add, Extension::kSPV_EXT_shader_atomic_float_min_max, Extension::kSPV_EXT_shader_image_int64, Extension::kSPV_EXT_shader_stencil_export, Extension::kSPV_EXT_shader_tile_image, Extension::kSPV_EXT_shader_viewport_index_layer, Extension::kSPV_GOOGLE_decorate_string, Extension::kSPV_GOOGLE_hlsl_functionality1, Extension::kSPV_GOOGLE_user_type, Extension::kSPV_INTEL_arbitrary_precision_fixed_point, Extension::kSPV_INTEL_arbitrary_precision_floating_point, Extension::kSPV_INTEL_arbitrary_precision_integers, Extension::kSPV_INTEL_bfloat16_conversion, Extension::kSPV_INTEL_blocking_pipes, Extension::kSPV_INTEL_cache_controls, Extension::kSPV_INTEL_debug_module, Extension::kSPV_INTEL_device_side_avc_motion_estimation, Extension::kSPV_INTEL_float_controls2, Extension::kSPV_INTEL_fp_fast_math_mode, Extension::kSPV_INTEL_fp_max_error, Extension::kSPV_INTEL_fpga_argument_interfaces, Extension::kSPV_INTEL_fpga_buffer_location, Extension::kSPV_INTEL_fpga_cluster_attributes, Extension::kSPV_INTEL_fpga_dsp_control, Extension::kSPV_INTEL_fpga_invocation_pipelining_attributes, Extension::kSPV_INTEL_fpga_latency_control, Extension::kSPV_INTEL_fpga_loop_controls, Extension::kSPV_INTEL_fpga_memory_accesses, Extension::kSPV_INTEL_fpga_memory_attributes, Extension::kSPV_INTEL_fpga_reg, Extension::kSPV_INTEL_function_pointers, Extension::kSPV_INTEL_global_variable_fpga_decorations, Extension::kSPV_INTEL_global_variable_host_access, Extension::kSPV_INTEL_inline_assembly, Extension::kSPV_INTEL_io_pipes, Extension::kSPV_INTEL_kernel_attributes, Extension::kSPV_INTEL_long_composites, Extension::kSPV_INTEL_loop_fuse, Extension::kSPV_INTEL_media_block_io, Extension::kSPV_INTEL_memory_access_aliasing, Extension::kSPV_INTEL_optnone, Extension::kSPV_INTEL_runtime_aligned, Extension::kSPV_INTEL_shader_integer_functions2, Extension::kSPV_INTEL_split_barrier, Extension::kSPV_INTEL_subgroups, Extension::kSPV_INTEL_unstructured_loop_controls, Extension::kSPV_INTEL_usm_storage_classes, Extension::kSPV_INTEL_variable_length_array, Extension::kSPV_INTEL_vector_compute, Extension::kSPV_KHR_16bit_storage, Extension::kSPV_KHR_8bit_storage, Extension::kSPV_KHR_bit_instructions, Extension::kSPV_KHR_cooperative_matrix, Extension::kSPV_KHR_device_group, Extension::kSPV_KHR_expect_assume, Extension::kSPV_KHR_float_controls, Extension::kSPV_KHR_fragment_shader_barycentric, Extension::kSPV_KHR_fragment_shading_rate, Extension::kSPV_KHR_integer_dot_product, Extension::kSPV_KHR_linkonce_odr, Extension::kSPV_KHR_multiview, Extension::kSPV_KHR_no_integer_wrap_decoration, Extension::kSPV_KHR_non_semantic_info, Extension::kSPV_KHR_physical_storage_buffer, Extension::kSPV_KHR_post_depth_coverage, Extension::kSPV_KHR_ray_cull_mask, Extension::kSPV_KHR_ray_query, Extension::kSPV_KHR_ray_tracing, Extension::kSPV_KHR_ray_tracing_position_fetch, Extension::kSPV_KHR_shader_atomic_counter_ops, Extension::kSPV_KHR_shader_ballot, Extension::kSPV_KHR_shader_clock, Extension::kSPV_KHR_shader_draw_parameters, Extension::kSPV_KHR_storage_buffer_storage_class, Extension::kSPV_KHR_subgroup_rotate, Extension::kSPV_KHR_subgroup_uniform_control_flow, Extension::kSPV_KHR_subgroup_vote, Extension::kSPV_KHR_terminate_invocation, Extension::kSPV_KHR_uniform_group_instructions, Extension::kSPV_KHR_variable_pointers, Extension::kSPV_KHR_vulkan_memory_model, Extension::kSPV_KHR_workgroup_memory_explicit_layout, Extension::kSPV_NVX_multiview_per_view_attributes, Extension::kSPV_NV_bindless_texture, Extension::kSPV_NV_compute_shader_derivatives, Extension::kSPV_NV_cooperative_matrix, Extension::kSPV_NV_displacement_micromap, Extension::kSPV_NV_fragment_shader_barycentric, Extension::kSPV_NV_geometry_shader_passthrough, Extension::kSPV_NV_mesh_shader, Extension::kSPV_NV_ray_tracing, Extension::kSPV_NV_ray_tracing_motion_blur, Extension::kSPV_NV_sample_mask_override_coverage, Extension::kSPV_NV_shader_image_footprint, Extension::kSPV_NV_shader_invocation_reorder, Extension::kSPV_NV_shader_sm_builtins, Extension::kSPV_NV_shader_subgroup_partitioned, Extension::kSPV_NV_shading_rate, Extension::kSPV_NV_stereo_view_rendering, Extension::kSPV_NV_viewport_array2, Extension::kSPV_QCOM_image_processing, Extension::kSPV_VALIDATOR_ignore_type_decl_unique }; + static const char* known_ext_strs[] = { "SPV_AMDX_shader_enqueue", "SPV_AMD_gcn_shader", "SPV_AMD_gpu_shader_half_float", "SPV_AMD_gpu_shader_half_float_fetch", "SPV_AMD_gpu_shader_int16", "SPV_AMD_shader_ballot", "SPV_AMD_shader_early_and_late_fragment_tests", "SPV_AMD_shader_explicit_vertex_parameter", "SPV_AMD_shader_fragment_mask", "SPV_AMD_shader_image_load_store_lod", "SPV_AMD_shader_trinary_minmax", "SPV_AMD_texture_gather_bias_lod", "SPV_ARM_core_builtins", "SPV_EXT_demote_to_helper_invocation", "SPV_EXT_descriptor_indexing", "SPV_EXT_fragment_fully_covered", "SPV_EXT_fragment_invocation_density", "SPV_EXT_fragment_shader_interlock", "SPV_EXT_mesh_shader", "SPV_EXT_opacity_micromap", "SPV_EXT_physical_storage_buffer", "SPV_EXT_shader_atomic_float16_add", "SPV_EXT_shader_atomic_float_add", "SPV_EXT_shader_atomic_float_min_max", "SPV_EXT_shader_image_int64", "SPV_EXT_shader_stencil_export", "SPV_EXT_shader_tile_image", "SPV_EXT_shader_viewport_index_layer", "SPV_GOOGLE_decorate_string", "SPV_GOOGLE_hlsl_functionality1", "SPV_GOOGLE_user_type", "SPV_INTEL_arbitrary_precision_fixed_point", "SPV_INTEL_arbitrary_precision_floating_point", "SPV_INTEL_arbitrary_precision_integers", "SPV_INTEL_bfloat16_conversion", "SPV_INTEL_blocking_pipes", "SPV_INTEL_cache_controls", "SPV_INTEL_debug_module", "SPV_INTEL_device_side_avc_motion_estimation", "SPV_INTEL_float_controls2", "SPV_INTEL_fp_fast_math_mode", "SPV_INTEL_fp_max_error", "SPV_INTEL_fpga_argument_interfaces", "SPV_INTEL_fpga_buffer_location", "SPV_INTEL_fpga_cluster_attributes", "SPV_INTEL_fpga_dsp_control", "SPV_INTEL_fpga_invocation_pipelining_attributes", "SPV_INTEL_fpga_latency_control", "SPV_INTEL_fpga_loop_controls", "SPV_INTEL_fpga_memory_accesses", "SPV_INTEL_fpga_memory_attributes", "SPV_INTEL_fpga_reg", "SPV_INTEL_function_pointers", "SPV_INTEL_global_variable_fpga_decorations", "SPV_INTEL_global_variable_host_access", "SPV_INTEL_inline_assembly", "SPV_INTEL_io_pipes", "SPV_INTEL_kernel_attributes", "SPV_INTEL_long_composites", "SPV_INTEL_loop_fuse", "SPV_INTEL_masked_gather_scatter", "SPV_INTEL_media_block_io", "SPV_INTEL_memory_access_aliasing", "SPV_INTEL_optnone", "SPV_INTEL_runtime_aligned", "SPV_INTEL_shader_integer_functions2", "SPV_INTEL_split_barrier", "SPV_INTEL_subgroups", "SPV_INTEL_unstructured_loop_controls", "SPV_INTEL_usm_storage_classes", "SPV_INTEL_variable_length_array", "SPV_INTEL_vector_compute", "SPV_KHR_16bit_storage", "SPV_KHR_8bit_storage", "SPV_KHR_bit_instructions", "SPV_KHR_cooperative_matrix", "SPV_KHR_device_group", "SPV_KHR_expect_assume", "SPV_KHR_float_controls", "SPV_KHR_float_controls2", "SPV_KHR_fragment_shader_barycentric", "SPV_KHR_fragment_shading_rate", "SPV_KHR_integer_dot_product", "SPV_KHR_linkonce_odr", "SPV_KHR_maximal_reconvergence", "SPV_KHR_multiview", "SPV_KHR_no_integer_wrap_decoration", "SPV_KHR_non_semantic_info", "SPV_KHR_physical_storage_buffer", "SPV_KHR_post_depth_coverage", "SPV_KHR_quad_control", "SPV_KHR_ray_cull_mask", "SPV_KHR_ray_query", "SPV_KHR_ray_tracing", "SPV_KHR_ray_tracing_position_fetch", "SPV_KHR_shader_atomic_counter_ops", "SPV_KHR_shader_ballot", "SPV_KHR_shader_clock", "SPV_KHR_shader_draw_parameters", "SPV_KHR_storage_buffer_storage_class", "SPV_KHR_subgroup_rotate", "SPV_KHR_subgroup_uniform_control_flow", "SPV_KHR_subgroup_vote", "SPV_KHR_terminate_invocation", "SPV_KHR_uniform_group_instructions", "SPV_KHR_variable_pointers", "SPV_KHR_vulkan_memory_model", "SPV_KHR_workgroup_memory_explicit_layout", "SPV_NVX_multiview_per_view_attributes", "SPV_NV_bindless_texture", "SPV_NV_compute_shader_derivatives", "SPV_NV_cooperative_matrix", "SPV_NV_displacement_micromap", "SPV_NV_fragment_shader_barycentric", "SPV_NV_geometry_shader_passthrough", "SPV_NV_mesh_shader", "SPV_NV_ray_tracing", "SPV_NV_ray_tracing_motion_blur", "SPV_NV_sample_mask_override_coverage", "SPV_NV_shader_image_footprint", "SPV_NV_shader_invocation_reorder", "SPV_NV_shader_sm_builtins", "SPV_NV_shader_subgroup_partitioned", "SPV_NV_shading_rate", "SPV_NV_stereo_view_rendering", "SPV_NV_viewport_array2", "SPV_QCOM_image_processing", "SPV_VALIDATOR_ignore_type_decl_unique" }; + static const Extension known_ext_ids[] = { Extension::kSPV_AMDX_shader_enqueue, Extension::kSPV_AMD_gcn_shader, Extension::kSPV_AMD_gpu_shader_half_float, Extension::kSPV_AMD_gpu_shader_half_float_fetch, Extension::kSPV_AMD_gpu_shader_int16, Extension::kSPV_AMD_shader_ballot, Extension::kSPV_AMD_shader_early_and_late_fragment_tests, Extension::kSPV_AMD_shader_explicit_vertex_parameter, Extension::kSPV_AMD_shader_fragment_mask, Extension::kSPV_AMD_shader_image_load_store_lod, Extension::kSPV_AMD_shader_trinary_minmax, Extension::kSPV_AMD_texture_gather_bias_lod, Extension::kSPV_ARM_core_builtins, Extension::kSPV_EXT_demote_to_helper_invocation, Extension::kSPV_EXT_descriptor_indexing, Extension::kSPV_EXT_fragment_fully_covered, Extension::kSPV_EXT_fragment_invocation_density, Extension::kSPV_EXT_fragment_shader_interlock, Extension::kSPV_EXT_mesh_shader, Extension::kSPV_EXT_opacity_micromap, Extension::kSPV_EXT_physical_storage_buffer, Extension::kSPV_EXT_shader_atomic_float16_add, Extension::kSPV_EXT_shader_atomic_float_add, Extension::kSPV_EXT_shader_atomic_float_min_max, Extension::kSPV_EXT_shader_image_int64, Extension::kSPV_EXT_shader_stencil_export, Extension::kSPV_EXT_shader_tile_image, Extension::kSPV_EXT_shader_viewport_index_layer, Extension::kSPV_GOOGLE_decorate_string, Extension::kSPV_GOOGLE_hlsl_functionality1, Extension::kSPV_GOOGLE_user_type, Extension::kSPV_INTEL_arbitrary_precision_fixed_point, Extension::kSPV_INTEL_arbitrary_precision_floating_point, Extension::kSPV_INTEL_arbitrary_precision_integers, Extension::kSPV_INTEL_bfloat16_conversion, Extension::kSPV_INTEL_blocking_pipes, Extension::kSPV_INTEL_cache_controls, Extension::kSPV_INTEL_debug_module, Extension::kSPV_INTEL_device_side_avc_motion_estimation, Extension::kSPV_INTEL_float_controls2, Extension::kSPV_INTEL_fp_fast_math_mode, Extension::kSPV_INTEL_fp_max_error, Extension::kSPV_INTEL_fpga_argument_interfaces, Extension::kSPV_INTEL_fpga_buffer_location, Extension::kSPV_INTEL_fpga_cluster_attributes, Extension::kSPV_INTEL_fpga_dsp_control, Extension::kSPV_INTEL_fpga_invocation_pipelining_attributes, Extension::kSPV_INTEL_fpga_latency_control, Extension::kSPV_INTEL_fpga_loop_controls, Extension::kSPV_INTEL_fpga_memory_accesses, Extension::kSPV_INTEL_fpga_memory_attributes, Extension::kSPV_INTEL_fpga_reg, Extension::kSPV_INTEL_function_pointers, Extension::kSPV_INTEL_global_variable_fpga_decorations, Extension::kSPV_INTEL_global_variable_host_access, Extension::kSPV_INTEL_inline_assembly, Extension::kSPV_INTEL_io_pipes, Extension::kSPV_INTEL_kernel_attributes, Extension::kSPV_INTEL_long_composites, Extension::kSPV_INTEL_loop_fuse, Extension::kSPV_INTEL_masked_gather_scatter, Extension::kSPV_INTEL_media_block_io, Extension::kSPV_INTEL_memory_access_aliasing, Extension::kSPV_INTEL_optnone, Extension::kSPV_INTEL_runtime_aligned, Extension::kSPV_INTEL_shader_integer_functions2, Extension::kSPV_INTEL_split_barrier, Extension::kSPV_INTEL_subgroups, Extension::kSPV_INTEL_unstructured_loop_controls, Extension::kSPV_INTEL_usm_storage_classes, Extension::kSPV_INTEL_variable_length_array, Extension::kSPV_INTEL_vector_compute, Extension::kSPV_KHR_16bit_storage, Extension::kSPV_KHR_8bit_storage, Extension::kSPV_KHR_bit_instructions, Extension::kSPV_KHR_cooperative_matrix, Extension::kSPV_KHR_device_group, Extension::kSPV_KHR_expect_assume, Extension::kSPV_KHR_float_controls, Extension::kSPV_KHR_float_controls2, Extension::kSPV_KHR_fragment_shader_barycentric, Extension::kSPV_KHR_fragment_shading_rate, Extension::kSPV_KHR_integer_dot_product, Extension::kSPV_KHR_linkonce_odr, Extension::kSPV_KHR_maximal_reconvergence, Extension::kSPV_KHR_multiview, Extension::kSPV_KHR_no_integer_wrap_decoration, Extension::kSPV_KHR_non_semantic_info, Extension::kSPV_KHR_physical_storage_buffer, Extension::kSPV_KHR_post_depth_coverage, Extension::kSPV_KHR_quad_control, Extension::kSPV_KHR_ray_cull_mask, Extension::kSPV_KHR_ray_query, Extension::kSPV_KHR_ray_tracing, Extension::kSPV_KHR_ray_tracing_position_fetch, Extension::kSPV_KHR_shader_atomic_counter_ops, Extension::kSPV_KHR_shader_ballot, Extension::kSPV_KHR_shader_clock, Extension::kSPV_KHR_shader_draw_parameters, Extension::kSPV_KHR_storage_buffer_storage_class, Extension::kSPV_KHR_subgroup_rotate, Extension::kSPV_KHR_subgroup_uniform_control_flow, Extension::kSPV_KHR_subgroup_vote, Extension::kSPV_KHR_terminate_invocation, Extension::kSPV_KHR_uniform_group_instructions, Extension::kSPV_KHR_variable_pointers, Extension::kSPV_KHR_vulkan_memory_model, Extension::kSPV_KHR_workgroup_memory_explicit_layout, Extension::kSPV_NVX_multiview_per_view_attributes, Extension::kSPV_NV_bindless_texture, Extension::kSPV_NV_compute_shader_derivatives, Extension::kSPV_NV_cooperative_matrix, Extension::kSPV_NV_displacement_micromap, Extension::kSPV_NV_fragment_shader_barycentric, Extension::kSPV_NV_geometry_shader_passthrough, Extension::kSPV_NV_mesh_shader, Extension::kSPV_NV_ray_tracing, Extension::kSPV_NV_ray_tracing_motion_blur, Extension::kSPV_NV_sample_mask_override_coverage, Extension::kSPV_NV_shader_image_footprint, Extension::kSPV_NV_shader_invocation_reorder, Extension::kSPV_NV_shader_sm_builtins, Extension::kSPV_NV_shader_subgroup_partitioned, Extension::kSPV_NV_shading_rate, Extension::kSPV_NV_stereo_view_rendering, Extension::kSPV_NV_viewport_array2, Extension::kSPV_QCOM_image_processing, Extension::kSPV_VALIDATOR_ignore_type_decl_unique }; const auto b = std::begin(known_ext_strs); const auto e = std::end(known_ext_strs); const auto found = std::equal_range( @@ -500,6 +508,8 @@ const char* CapabilityToString(spv::Capability capability) { return "ShaderClockKHR"; case spv::Capability::ShaderEnqueueAMDX: return "ShaderEnqueueAMDX"; + case spv::Capability::QuadControlKHR: + return "QuadControlKHR"; case spv::Capability::SampleMaskOverrideCoverageNV: return "SampleMaskOverrideCoverageNV"; case spv::Capability::GeometryShaderPassthroughNV: @@ -692,6 +702,8 @@ const char* CapabilityToString(spv::Capability capability) { return "BitInstructions"; case spv::Capability::GroupNonUniformRotateKHR: return "GroupNonUniformRotateKHR"; + case spv::Capability::FloatControls2: + return "FloatControls2"; case spv::Capability::AtomicFloat32AddEXT: return "AtomicFloat32AddEXT"; case spv::Capability::AtomicFloat64AddEXT: @@ -724,6 +736,8 @@ const char* CapabilityToString(spv::Capability capability) { return "GlobalVariableFPGADecorationsINTEL"; case spv::Capability::GroupUniformArithmeticKHR: return "GroupUniformArithmeticKHR"; + case spv::Capability::MaskedGatherScatterINTEL: + return "MaskedGatherScatterINTEL"; case spv::Capability::CacheControlsINTEL: return "CacheControlsINTEL"; case spv::Capability::Max: diff --git a/3rdparty/spirv-tools/include/generated/extension_enum.inc b/3rdparty/spirv-tools/include/generated/extension_enum.inc index 12699bcb8..07ffd9491 100644 --- a/3rdparty/spirv-tools/include/generated/extension_enum.inc +++ b/3rdparty/spirv-tools/include/generated/extension_enum.inc @@ -58,6 +58,7 @@ kSPV_INTEL_io_pipes, kSPV_INTEL_kernel_attributes, kSPV_INTEL_long_composites, kSPV_INTEL_loop_fuse, +kSPV_INTEL_masked_gather_scatter, kSPV_INTEL_media_block_io, kSPV_INTEL_memory_access_aliasing, kSPV_INTEL_optnone, @@ -76,15 +77,18 @@ kSPV_KHR_cooperative_matrix, kSPV_KHR_device_group, kSPV_KHR_expect_assume, kSPV_KHR_float_controls, +kSPV_KHR_float_controls2, kSPV_KHR_fragment_shader_barycentric, kSPV_KHR_fragment_shading_rate, kSPV_KHR_integer_dot_product, kSPV_KHR_linkonce_odr, +kSPV_KHR_maximal_reconvergence, kSPV_KHR_multiview, kSPV_KHR_no_integer_wrap_decoration, kSPV_KHR_non_semantic_info, kSPV_KHR_physical_storage_buffer, kSPV_KHR_post_depth_coverage, +kSPV_KHR_quad_control, kSPV_KHR_ray_cull_mask, kSPV_KHR_ray_query, kSPV_KHR_ray_tracing, diff --git a/3rdparty/spirv-tools/include/generated/generators.inc b/3rdparty/spirv-tools/include/generated/generators.inc index f70f13745..34f120f8d 100644 --- a/3rdparty/spirv-tools/include/generated/generators.inc +++ b/3rdparty/spirv-tools/include/generated/generators.inc @@ -38,4 +38,5 @@ {37, "heroseh", "Hero C Compiler", "heroseh Hero C Compiler"}, {38, "Meta", "SparkSL", "Meta SparkSL"}, {39, "SirLynix", "Nazara ShaderLang Compiler", "SirLynix Nazara ShaderLang Compiler"}, -{40, "NVIDIA", "Slang Compiler", "NVIDIA Slang Compiler"}, \ No newline at end of file +{40, "NVIDIA", "Slang Compiler", "NVIDIA Slang Compiler"}, +{41, "Zig Software Foundation", "Zig Compiler", "Zig Software Foundation Zig Compiler"}, \ No newline at end of file diff --git a/3rdparty/spirv-tools/include/generated/operand.kinds-unified1.inc b/3rdparty/spirv-tools/include/generated/operand.kinds-unified1.inc index 55aa97d19..3600f8833 100644 --- a/3rdparty/spirv-tools/include/generated/operand.kinds-unified1.inc +++ b/3rdparty/spirv-tools/include/generated/operand.kinds-unified1.inc @@ -15,7 +15,6 @@ static const spv::Capability pygen_variable_caps_DeviceEnqueue[] = {spv::Capabil static const spv::Capability pygen_variable_caps_DeviceGroup[] = {spv::Capability::DeviceGroup}; static const spv::Capability pygen_variable_caps_DrawParameters[] = {spv::Capability::DrawParameters}; static const spv::Capability pygen_variable_caps_DrawParametersMeshShadingNVMeshShadingEXT[] = {spv::Capability::DrawParameters, spv::Capability::MeshShadingNV, spv::Capability::MeshShadingEXT}; -static const spv::Capability pygen_variable_caps_FPFastMathModeINTEL[] = {spv::Capability::FPFastMathModeINTEL}; static const spv::Capability pygen_variable_caps_FPGAArgumentInterfacesINTEL[] = {spv::Capability::FPGAArgumentInterfacesINTEL}; static const spv::Capability pygen_variable_caps_FPGABufferLocationINTEL[] = {spv::Capability::FPGABufferLocationINTEL}; static const spv::Capability pygen_variable_caps_FPGAClusterAttributesINTEL[] = {spv::Capability::FPGAClusterAttributesINTEL}; @@ -29,6 +28,8 @@ static const spv::Capability pygen_variable_caps_FPGALoopControlsINTEL[] = {spv: static const spv::Capability pygen_variable_caps_FPGAMemoryAccessesINTEL[] = {spv::Capability::FPGAMemoryAccessesINTEL}; static const spv::Capability pygen_variable_caps_FPGAMemoryAttributesINTEL[] = {spv::Capability::FPGAMemoryAttributesINTEL}; static const spv::Capability pygen_variable_caps_FPMaxErrorINTEL[] = {spv::Capability::FPMaxErrorINTEL}; +static const spv::Capability pygen_variable_caps_FloatControls2[] = {spv::Capability::FloatControls2}; +static const spv::Capability pygen_variable_caps_FloatControls2FPFastMathModeINTEL[] = {spv::Capability::FloatControls2, spv::Capability::FPFastMathModeINTEL}; static const spv::Capability pygen_variable_caps_FragmentBarycentricNVFragmentBarycentricKHR[] = {spv::Capability::FragmentBarycentricNV, spv::Capability::FragmentBarycentricKHR}; static const spv::Capability pygen_variable_caps_FragmentDensityEXTShadingRateNV[] = {spv::Capability::FragmentDensityEXT, spv::Capability::ShadingRateNV}; static const spv::Capability pygen_variable_caps_FragmentFullyCoveredEXT[] = {spv::Capability::FragmentFullyCoveredEXT}; @@ -64,6 +65,7 @@ static const spv::Capability pygen_variable_caps_Int64[] = {spv::Capability::Int static const spv::Capability pygen_variable_caps_Int64ImageEXT[] = {spv::Capability::Int64ImageEXT}; static const spv::Capability pygen_variable_caps_Int8[] = {spv::Capability::Int8}; static const spv::Capability pygen_variable_caps_Kernel[] = {spv::Capability::Kernel}; +static const spv::Capability pygen_variable_caps_KernelFloatControls2[] = {spv::Capability::Kernel, spv::Capability::FloatControls2}; static const spv::Capability pygen_variable_caps_KernelGroupNonUniform[] = {spv::Capability::Kernel, spv::Capability::GroupNonUniform}; static const spv::Capability pygen_variable_caps_KernelGroupNonUniformSubgroupBallotKHR[] = {spv::Capability::Kernel, spv::Capability::GroupNonUniform, spv::Capability::SubgroupBallotKHR}; static const spv::Capability pygen_variable_caps_KernelGroupNonUniformArithmeticGroupNonUniformBallot[] = {spv::Capability::Kernel, spv::Capability::GroupNonUniformArithmetic, spv::Capability::GroupNonUniformBallot}; @@ -83,6 +85,7 @@ static const spv::Capability pygen_variable_caps_OptNoneINTEL[] = {spv::Capabili static const spv::Capability pygen_variable_caps_PerViewAttributesNVMeshShadingNV[] = {spv::Capability::PerViewAttributesNV, spv::Capability::MeshShadingNV}; static const spv::Capability pygen_variable_caps_PhysicalStorageBufferAddresses[] = {spv::Capability::PhysicalStorageBufferAddresses}; static const spv::Capability pygen_variable_caps_Pipes[] = {spv::Capability::Pipes}; +static const spv::Capability pygen_variable_caps_QuadControlKHR[] = {spv::Capability::QuadControlKHR}; static const spv::Capability pygen_variable_caps_RayCullMaskKHR[] = {spv::Capability::RayCullMaskKHR}; static const spv::Capability pygen_variable_caps_RayQueryKHR[] = {spv::Capability::RayQueryKHR}; static const spv::Capability pygen_variable_caps_RayQueryKHRRayTracingKHR[] = {spv::Capability::RayQueryKHR, spv::Capability::RayTracingKHR}; @@ -196,6 +199,7 @@ static const spvtools::Extension pygen_variable_exts_SPV_INTEL_io_pipes[] = {spv static const spvtools::Extension pygen_variable_exts_SPV_INTEL_kernel_attributes[] = {spvtools::Extension::kSPV_INTEL_kernel_attributes}; static const spvtools::Extension pygen_variable_exts_SPV_INTEL_long_composites[] = {spvtools::Extension::kSPV_INTEL_long_composites}; static const spvtools::Extension pygen_variable_exts_SPV_INTEL_loop_fuse[] = {spvtools::Extension::kSPV_INTEL_loop_fuse}; +static const spvtools::Extension pygen_variable_exts_SPV_INTEL_masked_gather_scatter[] = {spvtools::Extension::kSPV_INTEL_masked_gather_scatter}; static const spvtools::Extension pygen_variable_exts_SPV_INTEL_media_block_io[] = {spvtools::Extension::kSPV_INTEL_media_block_io}; static const spvtools::Extension pygen_variable_exts_SPV_INTEL_memory_access_aliasing[] = {spvtools::Extension::kSPV_INTEL_memory_access_aliasing}; static const spvtools::Extension pygen_variable_exts_SPV_INTEL_optnone[] = {spvtools::Extension::kSPV_INTEL_optnone}; @@ -214,13 +218,16 @@ static const spvtools::Extension pygen_variable_exts_SPV_KHR_cooperative_matrix[ static const spvtools::Extension pygen_variable_exts_SPV_KHR_device_group[] = {spvtools::Extension::kSPV_KHR_device_group}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_expect_assume[] = {spvtools::Extension::kSPV_KHR_expect_assume}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_float_controls[] = {spvtools::Extension::kSPV_KHR_float_controls}; +static const spvtools::Extension pygen_variable_exts_SPV_KHR_float_controls2[] = {spvtools::Extension::kSPV_KHR_float_controls2}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_fragment_shader_barycentricSPV_NV_fragment_shader_barycentric[] = {spvtools::Extension::kSPV_KHR_fragment_shader_barycentric, spvtools::Extension::kSPV_NV_fragment_shader_barycentric}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_fragment_shading_rate[] = {spvtools::Extension::kSPV_KHR_fragment_shading_rate}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_integer_dot_product[] = {spvtools::Extension::kSPV_KHR_integer_dot_product}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_linkonce_odr[] = {spvtools::Extension::kSPV_KHR_linkonce_odr}; +static const spvtools::Extension pygen_variable_exts_SPV_KHR_maximal_reconvergence[] = {spvtools::Extension::kSPV_KHR_maximal_reconvergence}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_multiview[] = {spvtools::Extension::kSPV_KHR_multiview}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_no_integer_wrap_decoration[] = {spvtools::Extension::kSPV_KHR_no_integer_wrap_decoration}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_post_depth_coverage[] = {spvtools::Extension::kSPV_KHR_post_depth_coverage}; +static const spvtools::Extension pygen_variable_exts_SPV_KHR_quad_control[] = {spvtools::Extension::kSPV_KHR_quad_control}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_ray_cull_mask[] = {spvtools::Extension::kSPV_KHR_ray_cull_mask}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_ray_query[] = {spvtools::Extension::kSPV_KHR_ray_query}; static const spvtools::Extension pygen_variable_exts_SPV_KHR_ray_querySPV_KHR_ray_tracing[] = {spvtools::Extension::kSPV_KHR_ray_query, spvtools::Extension::kSPV_KHR_ray_tracing}; @@ -290,8 +297,11 @@ static const spv_operand_desc_t pygen_variable_FPFastMathModeEntries[] = { {"NSZ", 0x0004, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"AllowRecip", 0x0008, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"Fast", 0x0010, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, - {"AllowContractFastINTEL", 0x10000, 1, pygen_variable_caps_FPFastMathModeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, - {"AllowReassocINTEL", 0x20000, 1, pygen_variable_caps_FPFastMathModeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu} + {"AllowContract", 0x10000, 2, pygen_variable_caps_FloatControls2FPFastMathModeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, + {"AllowContractFastINTEL", 0x10000, 2, pygen_variable_caps_FloatControls2FPFastMathModeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, + {"AllowReassoc", 0x20000, 2, pygen_variable_caps_FloatControls2FPFastMathModeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, + {"AllowReassocINTEL", 0x20000, 2, pygen_variable_caps_FloatControls2FPFastMathModeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, + {"AllowTransform", 0x40000, 1, pygen_variable_caps_FloatControls2, 0, nullptr, {}, 0xffffffffu, 0xffffffffu} }; static const spv_operand_desc_t pygen_variable_SelectionControlEntries[] = { @@ -408,7 +418,8 @@ static const spv_operand_desc_t pygen_variable_SourceLanguageEntries[] = { {"HERO_C", 8, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"NZSL", 9, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"WGSL", 10, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, - {"Slang", 11, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu} + {"Slang", 11, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, + {"Zig", 12, 0, nullptr, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu} }; static const spv_operand_desc_t pygen_variable_ExecutionModelEntries[] = { @@ -515,6 +526,8 @@ static const spv_operand_desc_t pygen_variable_ExecutionModeEntries[] = { {"StencilRefUnchangedBackAMD", 5082, 1, pygen_variable_caps_StencilExportEXT, 2, pygen_variable_exts_SPV_AMD_shader_early_and_late_fragment_testsSPV_EXT_shader_stencil_export, {}, 0xffffffffu, 0xffffffffu}, {"StencilRefGreaterBackAMD", 5083, 1, pygen_variable_caps_StencilExportEXT, 2, pygen_variable_exts_SPV_AMD_shader_early_and_late_fragment_testsSPV_EXT_shader_stencil_export, {}, 0xffffffffu, 0xffffffffu}, {"StencilRefLessBackAMD", 5084, 1, pygen_variable_caps_StencilExportEXT, 2, pygen_variable_exts_SPV_AMD_shader_early_and_late_fragment_testsSPV_EXT_shader_stencil_export, {}, 0xffffffffu, 0xffffffffu}, + {"QuadDerivativesKHR", 5088, 1, pygen_variable_caps_QuadControlKHR, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, + {"RequireFullQuadsKHR", 5089, 1, pygen_variable_caps_QuadControlKHR, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}, {"OutputLinesNV", 5269, 2, pygen_variable_caps_MeshShadingNVMeshShadingEXT, 2, pygen_variable_exts_SPV_EXT_mesh_shaderSPV_NV_mesh_shader, {}, 0xffffffffu, 0xffffffffu}, {"OutputLinesEXT", 5269, 2, pygen_variable_caps_MeshShadingNVMeshShadingEXT, 2, pygen_variable_exts_SPV_EXT_mesh_shaderSPV_NV_mesh_shader, {}, 0xffffffffu, 0xffffffffu}, {"OutputPrimitivesNV", 5270, 2, pygen_variable_caps_MeshShadingNVMeshShadingEXT, 2, pygen_variable_exts_SPV_EXT_mesh_shaderSPV_NV_mesh_shader, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu}, @@ -539,6 +552,8 @@ static const spv_operand_desc_t pygen_variable_ExecutionModeEntries[] = { {"NoGlobalOffsetINTEL", 5895, 1, pygen_variable_caps_KernelAttributesINTEL, 1, pygen_variable_exts_SPV_INTEL_kernel_attributes, {}, 0xffffffffu, 0xffffffffu}, {"NumSIMDWorkitemsINTEL", 5896, 1, pygen_variable_caps_FPGAKernelAttributesINTEL, 1, pygen_variable_exts_SPV_INTEL_kernel_attributes, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu}, {"SchedulerTargetFmaxMhzINTEL", 5903, 1, pygen_variable_caps_FPGAKernelAttributesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu}, + {"MaximallyReconvergesKHR", 6023, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_KHR_maximal_reconvergence, {}, 0xffffffffu, 0xffffffffu}, + {"FPFastMathDefault", 6028, 1, pygen_variable_caps_FloatControls2, 0, nullptr, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0xffffffffu, 0xffffffffu}, {"StreamingInterfaceINTEL", 6154, 1, pygen_variable_caps_FPGAKernelAttributesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu}, {"RegisterMapInterfaceINTEL", 6160, 1, pygen_variable_caps_FPGAKernelAttributesv2INTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu}, {"NamedBarrierCountINTEL", 6417, 1, pygen_variable_caps_VectorComputeINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu} @@ -802,7 +817,7 @@ static const spv_operand_desc_t pygen_variable_DecorationEntries[] = { {"XfbStride", 37, 1, pygen_variable_caps_TransformFeedback, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"FuncParamAttr", 38, 1, pygen_variable_caps_Kernel, 0, nullptr, {SPV_OPERAND_TYPE_FUNCTION_PARAMETER_ATTRIBUTE}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"FPRoundingMode", 39, 0, nullptr, 0, nullptr, {SPV_OPERAND_TYPE_FP_ROUNDING_MODE}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, - {"FPFastMathMode", 40, 1, pygen_variable_caps_Kernel, 0, nullptr, {SPV_OPERAND_TYPE_FP_FAST_MATH_MODE}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, + {"FPFastMathMode", 40, 2, pygen_variable_caps_KernelFloatControls2, 0, nullptr, {SPV_OPERAND_TYPE_FP_FAST_MATH_MODE}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"LinkageAttributes", 41, 1, pygen_variable_caps_Linkage, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_STRING, SPV_OPERAND_TYPE_LINKAGE_TYPE}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"NoContraction", 42, 1, pygen_variable_caps_Shader, 0, nullptr, {}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, {"InputAttachmentIndex", 43, 1, pygen_variable_caps_InputAttachment, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, SPV_SPIRV_VERSION_WORD(1,0), 0xffffffffu}, @@ -1194,6 +1209,7 @@ static const spv_operand_desc_t pygen_variable_CapabilityEntries[] = { {"Int64ImageEXT", 5016, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_EXT_shader_image_int64, {}, 0xffffffffu, 0xffffffffu}, {"ShaderClockKHR", 5055, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_shader_clock, {}, 0xffffffffu, 0xffffffffu}, {"ShaderEnqueueAMDX", 5067, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_AMDX_shader_enqueue, {}, 0xffffffffu, 0xffffffffu}, + {"QuadControlKHR", 5087, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_quad_control, {}, 0xffffffffu, 0xffffffffu}, {"SampleMaskOverrideCoverageNV", 5249, 1, pygen_variable_caps_SampleRateShading, 1, pygen_variable_exts_SPV_NV_sample_mask_override_coverage, {}, 0xffffffffu, 0xffffffffu}, {"GeometryShaderPassthroughNV", 5251, 1, pygen_variable_caps_Geometry, 1, pygen_variable_exts_SPV_NV_geometry_shader_passthrough, {}, 0xffffffffu, 0xffffffffu}, {"ShaderViewportIndexLayerEXT", 5254, 1, pygen_variable_caps_MultiViewport, 2, pygen_variable_exts_SPV_EXT_shader_viewport_index_layerSPV_NV_viewport_array2, {}, 0xffffffffu, 0xffffffffu}, @@ -1313,6 +1329,7 @@ static const spv_operand_desc_t pygen_variable_CapabilityEntries[] = { {"CooperativeMatrixKHR", 6022, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_cooperative_matrix, {}, 0xffffffffu, 0xffffffffu}, {"BitInstructions", 6025, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_bit_instructions, {}, 0xffffffffu, 0xffffffffu}, {"GroupNonUniformRotateKHR", 6026, 1, pygen_variable_caps_GroupNonUniform, 1, pygen_variable_exts_SPV_KHR_subgroup_rotate, {}, 0xffffffffu, 0xffffffffu}, + {"FloatControls2", 6029, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_float_controls2, {}, 0xffffffffu, 0xffffffffu}, {"AtomicFloat32AddEXT", 6033, 0, nullptr, 1, pygen_variable_exts_SPV_EXT_shader_atomic_float_add, {}, 0xffffffffu, 0xffffffffu}, {"AtomicFloat64AddEXT", 6034, 0, nullptr, 1, pygen_variable_exts_SPV_EXT_shader_atomic_float_add, {}, 0xffffffffu, 0xffffffffu}, {"LongCompositesINTEL", 6089, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_long_composites, {}, 0xffffffffu, 0xffffffffu}, @@ -1329,6 +1346,7 @@ static const spv_operand_desc_t pygen_variable_CapabilityEntries[] = { {"GlobalVariableHostAccessINTEL", 6187, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_global_variable_host_access, {}, 0xffffffffu, 0xffffffffu}, {"GlobalVariableFPGADecorationsINTEL", 6189, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_global_variable_fpga_decorations, {}, 0xffffffffu, 0xffffffffu}, {"GroupUniformArithmeticKHR", 6400, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_uniform_group_instructions, {}, 0xffffffffu, 0xffffffffu}, + {"MaskedGatherScatterINTEL", 6427, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_masked_gather_scatter, {}, 0xffffffffu, 0xffffffffu}, {"CacheControlsINTEL", 6441, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_cache_controls, {}, 0xffffffffu, 0xffffffffu} }; diff --git a/3rdparty/spirv-tools/include/spirv-tools/optimizer.hpp b/3rdparty/spirv-tools/include/spirv-tools/optimizer.hpp index bc4f5bee8..926e438fe 100644 --- a/3rdparty/spirv-tools/include/spirv-tools/optimizer.hpp +++ b/3rdparty/spirv-tools/include/spirv-tools/optimizer.hpp @@ -1003,6 +1003,10 @@ Optimizer::PassToken CreateSwitchDescriptorSetPass(uint32_t ds_from, // OpBeginInterlockInvocationEXT and one OpEndInterlockInvocationEXT, in that // order. Optimizer::PassToken CreateInvocationInterlockPlacementPass(); + +// Creates a pass to add/remove maximal reconvergence execution mode. +// This pass either adds or removes maximal reconvergence from all entry points. +Optimizer::PassToken CreateModifyMaximalReconvergencePass(bool add); } // namespace spvtools #endif // INCLUDE_SPIRV_TOOLS_OPTIMIZER_HPP_ diff --git a/3rdparty/spirv-tools/source/opcode.cpp b/3rdparty/spirv-tools/source/opcode.cpp index ffbb2e8ba..38d1a1be6 100644 --- a/3rdparty/spirv-tools/source/opcode.cpp +++ b/3rdparty/spirv-tools/source/opcode.cpp @@ -534,6 +534,8 @@ bool spvOpcodeIsNonUniformGroupOperation(spv::Op opcode) { case spv::Op::OpGroupNonUniformQuadBroadcast: case spv::Op::OpGroupNonUniformQuadSwap: case spv::Op::OpGroupNonUniformRotateKHR: + case spv::Op::OpGroupNonUniformQuadAllKHR: + case spv::Op::OpGroupNonUniformQuadAnyKHR: return true; default: return false; diff --git a/3rdparty/spirv-tools/source/opt/aggressive_dead_code_elim_pass.cpp b/3rdparty/spirv-tools/source/opt/aggressive_dead_code_elim_pass.cpp index b372571f5..4737da5f9 100644 --- a/3rdparty/spirv-tools/source/opt/aggressive_dead_code_elim_pass.cpp +++ b/3rdparty/spirv-tools/source/opt/aggressive_dead_code_elim_pass.cpp @@ -985,6 +985,7 @@ void AggressiveDCEPass::InitExtensions() { "SPV_NV_shader_image_footprint", "SPV_NV_shading_rate", "SPV_NV_mesh_shader", + "SPV_EXT_mesh_shader", "SPV_NV_ray_tracing", "SPV_KHR_ray_tracing", "SPV_KHR_ray_query", diff --git a/3rdparty/spirv-tools/source/opt/block_merge_util.cpp b/3rdparty/spirv-tools/source/opt/block_merge_util.cpp index fe23e36f9..42f695f23 100644 --- a/3rdparty/spirv-tools/source/opt/block_merge_util.cpp +++ b/3rdparty/spirv-tools/source/opt/block_merge_util.cpp @@ -98,6 +98,17 @@ bool CanMergeWithSuccessor(IRContext* context, BasicBlock* block) { return false; } + // Note: This means that the instructions in a break block will execute as if + // they were still diverged according to the loop iteration. This restricts + // potential transformations an implementation may perform on the IR to match + // shader author expectations. Similarly, instructions in the loop construct + // cannot be moved into the continue construct unless it can be proven that + // invocations are always converged. + if (succ_is_merge && context->get_feature_mgr()->HasExtension( + kSPV_KHR_maximal_reconvergence)) { + return false; + } + if (pred_is_merge && IsContinue(context, lab_id)) { // Cannot merge a continue target with a merge block. return false; diff --git a/3rdparty/spirv-tools/source/opt/const_folding_rules.cpp b/3rdparty/spirv-tools/source/opt/const_folding_rules.cpp index e676974c8..79f34acd3 100644 --- a/3rdparty/spirv-tools/source/opt/const_folding_rules.cpp +++ b/3rdparty/spirv-tools/source/opt/const_folding_rules.cpp @@ -21,6 +21,59 @@ namespace opt { namespace { constexpr uint32_t kExtractCompositeIdInIdx = 0; +// Returns the value obtained by extracting the |number_of_bits| least +// significant bits from |value|, and sign-extending it to 64-bits. +uint64_t SignExtendValue(uint64_t value, uint32_t number_of_bits) { + if (number_of_bits == 64) return value; + + uint64_t mask_for_sign_bit = 1ull << (number_of_bits - 1); + uint64_t mask_for_significant_bits = (mask_for_sign_bit << 1) - 1ull; + if (value & mask_for_sign_bit) { + // Set upper bits to 1 + value |= ~mask_for_significant_bits; + } else { + // Clear the upper bits + value &= mask_for_significant_bits; + } + return value; +} + +// Returns the value obtained by extracting the |number_of_bits| least +// significant bits from |value|, and zero-extending it to 64-bits. +uint64_t ZeroExtendValue(uint64_t value, uint32_t number_of_bits) { + if (number_of_bits == 64) return value; + + uint64_t mask_for_first_bit_to_clear = 1ull << (number_of_bits); + uint64_t mask_for_bits_to_keep = mask_for_first_bit_to_clear - 1; + value &= mask_for_bits_to_keep; + return value; +} + +// Returns a constant whose value is `value` and type is `type`. This constant +// will be generated by `const_mgr`. The type must be a scalar integer type. +const analysis::Constant* GenerateIntegerConstant( + const analysis::Integer* integer_type, uint64_t result, + analysis::ConstantManager* const_mgr) { + assert(integer_type != nullptr); + + std::vector words; + if (integer_type->width() == 64) { + // In the 64-bit case, two words are needed to represent the value. + words = {static_cast(result), + static_cast(result >> 32)}; + } else { + // In all other cases, only a single word is needed. + assert(integer_type->width() <= 32); + if (integer_type->IsSigned()) { + result = SignExtendValue(result, integer_type->width()); + } else { + result = ZeroExtendValue(result, integer_type->width()); + } + words = {static_cast(result)}; + } + return const_mgr->GetConstant(integer_type, words); +} + // Returns a constants with the value NaN of the given type. Only works for // 32-bit and 64-bit float point types. Returns |nullptr| if an error occurs. const analysis::Constant* GetNan(const analysis::Type* type, @@ -676,7 +729,6 @@ ConstantFoldingRule FoldUnaryOp(UnaryScalarFoldingRule scalar_rule) { return [scalar_rule](IRContext* context, Instruction* inst, const std::vector& constants) -> const analysis::Constant* { - analysis::ConstantManager* const_mgr = context->get_constant_mgr(); analysis::TypeManager* type_mgr = context->get_type_mgr(); const analysis::Type* result_type = type_mgr->GetType(inst->type_id()); @@ -716,6 +768,64 @@ ConstantFoldingRule FoldUnaryOp(UnaryScalarFoldingRule scalar_rule) { }; } +// Returns a |ConstantFoldingRule| that folds binary scalar ops +// using |scalar_rule| and binary vectors ops by applying +// |scalar_rule| to the elements of the vector. The folding rule assumes that op +// has two inputs. For regular instruction, those are in operands 0 and 1. For +// extended instruction, they are in operands 1 and 2. If an element in +// |constants| is not nullprt, then the constant's type is |Float|, |Integer|, +// or |Vector| whose element type is |Float| or |Integer|. +ConstantFoldingRule FoldBinaryOp(BinaryScalarFoldingRule scalar_rule) { + return [scalar_rule](IRContext* context, Instruction* inst, + const std::vector& constants) + -> const analysis::Constant* { + assert(constants.size() == inst->NumInOperands()); + assert(constants.size() == (inst->opcode() == spv::Op::OpExtInst ? 3 : 2)); + analysis::ConstantManager* const_mgr = context->get_constant_mgr(); + analysis::TypeManager* type_mgr = context->get_type_mgr(); + const analysis::Type* result_type = type_mgr->GetType(inst->type_id()); + const analysis::Vector* vector_type = result_type->AsVector(); + + const analysis::Constant* arg1 = + (inst->opcode() == spv::Op::OpExtInst) ? constants[1] : constants[0]; + const analysis::Constant* arg2 = + (inst->opcode() == spv::Op::OpExtInst) ? constants[2] : constants[1]; + + if (arg1 == nullptr || arg2 == nullptr) { + return nullptr; + } + + if (vector_type == nullptr) { + return scalar_rule(result_type, arg1, arg2, const_mgr); + } + + std::vector a_components; + std::vector b_components; + std::vector results_components; + + a_components = arg1->GetVectorComponents(const_mgr); + b_components = arg2->GetVectorComponents(const_mgr); + assert(a_components.size() == b_components.size()); + + // Fold each component of the vector. + for (uint32_t i = 0; i < a_components.size(); ++i) { + results_components.push_back(scalar_rule(vector_type->element_type(), + a_components[i], b_components[i], + const_mgr)); + if (results_components[i] == nullptr) { + return nullptr; + } + } + + // Build the constant object and return it. + std::vector ids; + for (const analysis::Constant* member : results_components) { + ids.push_back(const_mgr->GetDefiningInstruction(member)->result_id()); + } + return const_mgr->GetConstant(vector_type, ids); + }; +} + // Returns a |ConstantFoldingRule| that folds unary floating point scalar ops // using |scalar_rule| and unary float point vectors ops by applying // |scalar_rule| to the elements of the vector. The |ConstantFoldingRule| @@ -1587,6 +1697,72 @@ BinaryScalarFoldingRule FoldFTranscendentalBinary(double (*fp)(double, return nullptr; }; } + +enum Sign { Signed, Unsigned }; + +// Returns a BinaryScalarFoldingRule that applies `op` to the scalars. +// The `signedness` is used to determine if the operands should be interpreted +// as signed or unsigned. If the operands are signed, the value will be sign +// extended before the value is passed to `op`. Otherwise the values will be +// zero extended. +template +BinaryScalarFoldingRule FoldBinaryIntegerOperation(uint64_t (*op)(uint64_t, + uint64_t)) { + return + [op](const analysis::Type* result_type, const analysis::Constant* a, + const analysis::Constant* b, + analysis::ConstantManager* const_mgr) -> const analysis::Constant* { + assert(result_type != nullptr && a != nullptr && b != nullptr); + const analysis::Integer* integer_type = result_type->AsInteger(); + assert(integer_type != nullptr); + assert(integer_type == a->type()->AsInteger()); + assert(integer_type == b->type()->AsInteger()); + + // In SPIR-V, all operations support unsigned types, but the way they + // are interpreted depends on the opcode. This is why we use the + // template argument to determine how to interpret the operands. + uint64_t ia = (signedness == Signed ? a->GetSignExtendedValue() + : a->GetZeroExtendedValue()); + uint64_t ib = (signedness == Signed ? b->GetSignExtendedValue() + : b->GetZeroExtendedValue()); + uint64_t result = op(ia, ib); + + const analysis::Constant* result_constant = + GenerateIntegerConstant(integer_type, result, const_mgr); + return result_constant; + }; +} + +// A scalar folding rule that folds OpSConvert. +const analysis::Constant* FoldScalarSConvert( + const analysis::Type* result_type, const analysis::Constant* a, + analysis::ConstantManager* const_mgr) { + assert(result_type != nullptr); + assert(a != nullptr); + assert(const_mgr != nullptr); + const analysis::Integer* integer_type = result_type->AsInteger(); + assert(integer_type && "The result type of an SConvert"); + int64_t value = a->GetSignExtendedValue(); + return GenerateIntegerConstant(integer_type, value, const_mgr); +} + +// A scalar folding rule that folds OpUConvert. +const analysis::Constant* FoldScalarUConvert( + const analysis::Type* result_type, const analysis::Constant* a, + analysis::ConstantManager* const_mgr) { + assert(result_type != nullptr); + assert(a != nullptr); + assert(const_mgr != nullptr); + const analysis::Integer* integer_type = result_type->AsInteger(); + assert(integer_type && "The result type of an UConvert"); + uint64_t value = a->GetZeroExtendedValue(); + + // If the operand was an unsigned value with less than 32-bit, it would have + // been sign extended earlier, and we need to clear those bits. + auto* operand_type = a->type()->AsInteger(); + value = ZeroExtendValue(value, operand_type->width()); + return GenerateIntegerConstant(integer_type, value, const_mgr); +} } // namespace void ConstantFoldingRules::AddFoldingRules() { @@ -1604,6 +1780,8 @@ void ConstantFoldingRules::AddFoldingRules() { rules_[spv::Op::OpConvertFToU].push_back(FoldFToI()); rules_[spv::Op::OpConvertSToF].push_back(FoldIToF()); rules_[spv::Op::OpConvertUToF].push_back(FoldIToF()); + rules_[spv::Op::OpSConvert].push_back(FoldUnaryOp(FoldScalarSConvert)); + rules_[spv::Op::OpUConvert].push_back(FoldUnaryOp(FoldScalarUConvert)); rules_[spv::Op::OpDot].push_back(FoldOpDotWithConstants()); rules_[spv::Op::OpFAdd].push_back(FoldFAdd()); @@ -1662,6 +1840,46 @@ void ConstantFoldingRules::AddFoldingRules() { rules_[spv::Op::OpSNegate].push_back(FoldSNegate()); rules_[spv::Op::OpQuantizeToF16].push_back(FoldQuantizeToF16()); + rules_[spv::Op::OpIAdd].push_back( + FoldBinaryOp(FoldBinaryIntegerOperation( + [](uint64_t a, uint64_t b) { return a + b; }))); + rules_[spv::Op::OpISub].push_back( + FoldBinaryOp(FoldBinaryIntegerOperation( + [](uint64_t a, uint64_t b) { return a - b; }))); + rules_[spv::Op::OpIMul].push_back( + FoldBinaryOp(FoldBinaryIntegerOperation( + [](uint64_t a, uint64_t b) { return a * b; }))); + rules_[spv::Op::OpUDiv].push_back( + FoldBinaryOp(FoldBinaryIntegerOperation( + [](uint64_t a, uint64_t b) { return (b != 0 ? a / b : 0); }))); + rules_[spv::Op::OpSDiv].push_back(FoldBinaryOp( + FoldBinaryIntegerOperation([](uint64_t a, uint64_t b) { + return (b != 0 ? static_cast(static_cast(a) / + static_cast(b)) + : 0); + }))); + rules_[spv::Op::OpUMod].push_back( + FoldBinaryOp(FoldBinaryIntegerOperation( + [](uint64_t a, uint64_t b) { return (b != 0 ? a % b : 0); }))); + + rules_[spv::Op::OpSRem].push_back(FoldBinaryOp( + FoldBinaryIntegerOperation([](uint64_t a, uint64_t b) { + return (b != 0 ? static_cast(static_cast(a) % + static_cast(b)) + : 0); + }))); + + rules_[spv::Op::OpSMod].push_back(FoldBinaryOp( + FoldBinaryIntegerOperation([](uint64_t a, uint64_t b) { + if (b == 0) return static_cast(0ull); + + int64_t signed_a = static_cast(a); + int64_t signed_b = static_cast(b); + int64_t result = signed_a % signed_b; + if ((signed_b < 0) != (result < 0)) result += signed_b; + return static_cast(result); + }))); + // Add rules for GLSLstd450 FeatureManager* feature_manager = context_->get_feature_mgr(); uint32_t ext_inst_glslstd450_id = diff --git a/3rdparty/spirv-tools/source/opt/convert_to_half_pass.cpp b/3rdparty/spirv-tools/source/opt/convert_to_half_pass.cpp index cb0065d2d..e243bedf0 100644 --- a/3rdparty/spirv-tools/source/opt/convert_to_half_pass.cpp +++ b/3rdparty/spirv-tools/source/opt/convert_to_half_pass.cpp @@ -171,6 +171,19 @@ bool ConvertToHalfPass::RemoveRelaxedDecoration(uint32_t id) { bool ConvertToHalfPass::GenHalfArith(Instruction* inst) { bool modified = false; + // If this is a OpCompositeExtract instruction and has a struct operand, we + // should not relax this instruction. Doing so could cause a mismatch between + // the result type and the struct member type. + bool hasStructOperand = false; + if (inst->opcode() == spv::Op::OpCompositeExtract) { + inst->ForEachInId([&hasStructOperand, this](uint32_t* idp) { + Instruction* op_inst = get_def_use_mgr()->GetDef(*idp); + if (IsStruct(op_inst)) hasStructOperand = true; + }); + if (hasStructOperand) { + return false; + } + } // Convert all float32 based operands to float16 equivalent and change // instruction type to float16 equivalent. inst->ForEachInId([&inst, &modified, this](uint32_t* idp) { @@ -303,12 +316,19 @@ bool ConvertToHalfPass::CloseRelaxInst(Instruction* inst) { if (closure_ops_.count(inst->opcode()) == 0) return false; // Can relax if all float operands are relaxed bool relax = true; - inst->ForEachInId([&relax, this](uint32_t* idp) { + bool hasStructOperand = false; + inst->ForEachInId([&relax, &hasStructOperand, this](uint32_t* idp) { Instruction* op_inst = get_def_use_mgr()->GetDef(*idp); - if (IsStruct(op_inst)) relax = false; + if (IsStruct(op_inst)) hasStructOperand = true; if (!IsFloat(op_inst, 32)) return; if (!IsRelaxed(*idp)) relax = false; }); + // If the instruction has a struct operand, we should not relax it, even if + // all its uses are relaxed. Doing so could cause a mismatch between the + // result type and the struct member type. + if (hasStructOperand) { + return false; + } if (relax) { AddRelaxed(inst->result_id()); return true; diff --git a/3rdparty/spirv-tools/source/opt/def_use_manager.h b/3rdparty/spirv-tools/source/opt/def_use_manager.h index a8dbbc60b..13cf9bd3e 100644 --- a/3rdparty/spirv-tools/source/opt/def_use_manager.h +++ b/3rdparty/spirv-tools/source/opt/def_use_manager.h @@ -27,28 +27,6 @@ namespace spvtools { namespace opt { namespace analysis { -// Class for representing a use of id. Note that: -// * Result type id is a use. -// * Ids referenced in OpSectionMerge & OpLoopMerge are considered as use. -// * Ids referenced in OpPhi's in operands are considered as use. -struct Use { - Instruction* inst; // Instruction using the id. - uint32_t operand_index; // logical operand index of the id use. This can be - // the index of result type id. -}; - -inline bool operator==(const Use& lhs, const Use& rhs) { - return lhs.inst == rhs.inst && lhs.operand_index == rhs.operand_index; -} - -inline bool operator!=(const Use& lhs, const Use& rhs) { return !(lhs == rhs); } - -inline bool operator<(const Use& lhs, const Use& rhs) { - if (lhs.inst < rhs.inst) return true; - if (lhs.inst > rhs.inst) return false; - return lhs.operand_index < rhs.operand_index; -} - // Definition should never be null. User can be null, however, such an entry // should be used only for searching (e.g. all users of a particular definition) // and never stored in a container. diff --git a/3rdparty/spirv-tools/source/opt/desc_sroa.cpp b/3rdparty/spirv-tools/source/opt/desc_sroa.cpp index 8da0c864f..2c0f4829f 100644 --- a/3rdparty/spirv-tools/source/opt/desc_sroa.cpp +++ b/3rdparty/spirv-tools/source/opt/desc_sroa.cpp @@ -54,9 +54,10 @@ Pass::Status DescriptorScalarReplacement::Process() { bool DescriptorScalarReplacement::ReplaceCandidate(Instruction* var) { std::vector access_chain_work_list; std::vector load_work_list; + std::vector entry_point_work_list; bool failed = !get_def_use_mgr()->WhileEachUser( - var->result_id(), - [this, &access_chain_work_list, &load_work_list](Instruction* use) { + var->result_id(), [this, &access_chain_work_list, &load_work_list, + &entry_point_work_list](Instruction* use) { if (use->opcode() == spv::Op::OpName) { return true; } @@ -73,6 +74,9 @@ bool DescriptorScalarReplacement::ReplaceCandidate(Instruction* var) { case spv::Op::OpLoad: load_work_list.push_back(use); return true; + case spv::Op::OpEntryPoint: + entry_point_work_list.push_back(use); + return true; default: context()->EmitErrorMessage( "Variable cannot be replaced: invalid instruction", use); @@ -95,6 +99,11 @@ bool DescriptorScalarReplacement::ReplaceCandidate(Instruction* var) { return false; } } + for (Instruction* use : entry_point_work_list) { + if (!ReplaceEntryPoint(var, use)) { + return false; + } + } return true; } @@ -147,6 +156,42 @@ bool DescriptorScalarReplacement::ReplaceAccessChain(Instruction* var, return true; } +bool DescriptorScalarReplacement::ReplaceEntryPoint(Instruction* var, + Instruction* use) { + // Build a new |OperandList| for |use| that removes |var| and adds its + // replacement variables. + Instruction::OperandList new_operands; + + // Copy all operands except |var|. + bool found = false; + for (uint32_t idx = 0; idx < use->NumOperands(); idx++) { + Operand& op = use->GetOperand(idx); + if (op.type == SPV_OPERAND_TYPE_ID && op.words[0] == var->result_id()) { + found = true; + } else { + new_operands.emplace_back(op); + } + } + + if (!found) { + context()->EmitErrorMessage( + "Variable cannot be replaced: invalid instruction", use); + return false; + } + + // Add all new replacement variables. + uint32_t num_replacement_vars = + descsroautil::GetNumberOfElementsForArrayOrStruct(context(), var); + for (uint32_t i = 0; i < num_replacement_vars; i++) { + new_operands.push_back( + {SPV_OPERAND_TYPE_ID, {GetReplacementVariable(var, i)}}); + } + + use->ReplaceOperands(new_operands); + context()->UpdateDefUse(use); + return true; +} + uint32_t DescriptorScalarReplacement::GetReplacementVariable(Instruction* var, uint32_t idx) { auto replacement_vars = replacement_variables_.find(var); diff --git a/3rdparty/spirv-tools/source/opt/desc_sroa.h b/3rdparty/spirv-tools/source/opt/desc_sroa.h index 6a24fd871..901be3e98 100644 --- a/3rdparty/spirv-tools/source/opt/desc_sroa.h +++ b/3rdparty/spirv-tools/source/opt/desc_sroa.h @@ -64,6 +64,11 @@ class DescriptorScalarReplacement : public Pass { // otherwise. bool ReplaceLoadedValue(Instruction* var, Instruction* value); + // Replaces the given composite variable |var| in the OpEntryPoint with the + // new replacement variables, one for each element of the array |var|. Returns + // |true| if successful, and |false| otherwise. + bool ReplaceEntryPoint(Instruction* var, Instruction* use); + // Replaces the given OpCompositeExtract |extract| and all of its references // with an OpLoad of a replacement variable. |var| is the variable with // composite type whose value is being used by |extract|. Assumes that diff --git a/3rdparty/spirv-tools/source/opt/ir_context.h b/3rdparty/spirv-tools/source/opt/ir_context.h index de3c41066..5685db809 100644 --- a/3rdparty/spirv-tools/source/opt/ir_context.h +++ b/3rdparty/spirv-tools/source/opt/ir_context.h @@ -229,6 +229,8 @@ class IRContext { inline void AddExtInstImport(std::unique_ptr&& e); // Set the memory model for this module. inline void SetMemoryModel(std::unique_ptr&& m); + // Get the memory model for this module. + inline const Instruction* GetMemoryModel() const; // Appends an entry point instruction to this module. inline void AddEntryPoint(std::unique_ptr&& e); // Appends an execution mode instruction to this module. @@ -1156,6 +1158,10 @@ void IRContext::SetMemoryModel(std::unique_ptr&& m) { module()->SetMemoryModel(std::move(m)); } +const Instruction* IRContext::GetMemoryModel() const { + return module()->GetMemoryModel(); +} + void IRContext::AddEntryPoint(std::unique_ptr&& e) { module()->AddEntryPoint(std::move(e)); } diff --git a/3rdparty/spirv-tools/source/opt/local_access_chain_convert_pass.cpp b/3rdparty/spirv-tools/source/opt/local_access_chain_convert_pass.cpp index ea1bdeeb3..7ba75cb7a 100644 --- a/3rdparty/spirv-tools/source/opt/local_access_chain_convert_pass.cpp +++ b/3rdparty/spirv-tools/source/opt/local_access_chain_convert_pass.cpp @@ -420,8 +420,8 @@ void LocalAccessChainConvertPass::InitExtensions() { "SPV_EXT_demote_to_helper_invocation", "SPV_EXT_descriptor_indexing", "SPV_NV_fragment_shader_barycentric", "SPV_NV_compute_shader_derivatives", "SPV_NV_shader_image_footprint", - "SPV_NV_shading_rate", "SPV_NV_mesh_shader", "SPV_NV_ray_tracing", - "SPV_KHR_ray_tracing", "SPV_KHR_ray_query", + "SPV_NV_shading_rate", "SPV_NV_mesh_shader", "SPV_EXT_mesh_shader", + "SPV_NV_ray_tracing", "SPV_KHR_ray_tracing", "SPV_KHR_ray_query", "SPV_EXT_fragment_invocation_density", "SPV_KHR_terminate_invocation", "SPV_KHR_subgroup_uniform_control_flow", "SPV_KHR_integer_dot_product", "SPV_EXT_shader_image_int64", "SPV_KHR_non_semantic_info", diff --git a/3rdparty/spirv-tools/source/opt/local_single_block_elim_pass.cpp b/3rdparty/spirv-tools/source/opt/local_single_block_elim_pass.cpp index 7502d0497..d7a9295e8 100644 --- a/3rdparty/spirv-tools/source/opt/local_single_block_elim_pass.cpp +++ b/3rdparty/spirv-tools/source/opt/local_single_block_elim_pass.cpp @@ -273,6 +273,7 @@ void LocalSingleBlockLoadStoreElimPass::InitExtensions() { "SPV_NV_shader_image_footprint", "SPV_NV_shading_rate", "SPV_NV_mesh_shader", + "SPV_EXT_mesh_shader", "SPV_NV_ray_tracing", "SPV_KHR_ray_tracing", "SPV_KHR_ray_query", diff --git a/3rdparty/spirv-tools/source/opt/local_single_store_elim_pass.cpp b/3rdparty/spirv-tools/source/opt/local_single_store_elim_pass.cpp index f6fc2760e..7cd6b0eb4 100644 --- a/3rdparty/spirv-tools/source/opt/local_single_store_elim_pass.cpp +++ b/3rdparty/spirv-tools/source/opt/local_single_store_elim_pass.cpp @@ -124,6 +124,7 @@ void LocalSingleStoreElimPass::InitExtensionAllowList() { "SPV_NV_shader_image_footprint", "SPV_NV_shading_rate", "SPV_NV_mesh_shader", + "SPV_EXT_mesh_shader", "SPV_NV_ray_tracing", "SPV_KHR_ray_query", "SPV_EXT_fragment_invocation_density", diff --git a/3rdparty/spirv-tools/source/opt/modify_maximal_reconvergence.cpp b/3rdparty/spirv-tools/source/opt/modify_maximal_reconvergence.cpp new file mode 100644 index 000000000..dd79b6283 --- /dev/null +++ b/3rdparty/spirv-tools/source/opt/modify_maximal_reconvergence.cpp @@ -0,0 +1,103 @@ +// Copyright (c) 2024 Google LLC +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "modify_maximal_reconvergence.h" + +#include "source/opt/ir_context.h" +#include "source/util/make_unique.h" + +namespace spvtools { +namespace opt { + +Pass::Status ModifyMaximalReconvergence::Process() { + bool changed = false; + if (add_) { + changed = AddMaximalReconvergence(); + } else { + changed = RemoveMaximalReconvergence(); + } + return changed ? Pass::Status::SuccessWithChange + : Pass::Status::SuccessWithoutChange; +} + +bool ModifyMaximalReconvergence::AddMaximalReconvergence() { + bool changed = false; + bool has_extension = false; + bool has_shader = + context()->get_feature_mgr()->HasCapability(spv::Capability::Shader); + for (auto extension : context()->extensions()) { + if (extension.GetOperand(0).AsString() == "SPV_KHR_maximal_reconvergence") { + has_extension = true; + break; + } + } + + std::unordered_set entry_points_with_mode; + for (auto mode : get_module()->execution_modes()) { + if (spv::ExecutionMode(mode.GetSingleWordInOperand(1)) == + spv::ExecutionMode::MaximallyReconvergesKHR) { + entry_points_with_mode.insert(mode.GetSingleWordInOperand(0)); + } + } + + for (auto entry_point : get_module()->entry_points()) { + const uint32_t id = entry_point.GetSingleWordInOperand(1); + if (!entry_points_with_mode.count(id)) { + changed = true; + if (!has_extension) { + context()->AddExtension("SPV_KHR_maximal_reconvergence"); + has_extension = true; + } + if (!has_shader) { + context()->AddCapability(spv::Capability::Shader); + has_shader = true; + } + context()->AddExecutionMode(MakeUnique( + context(), spv::Op::OpExecutionMode, 0, 0, + std::initializer_list{ + {SPV_OPERAND_TYPE_ID, {id}}, + {SPV_OPERAND_TYPE_EXECUTION_MODE, + {static_cast( + spv::ExecutionMode::MaximallyReconvergesKHR)}}})); + entry_points_with_mode.insert(id); + } + } + + return changed; +} + +bool ModifyMaximalReconvergence::RemoveMaximalReconvergence() { + bool changed = false; + std::vector to_remove; + Instruction* mode = &*get_module()->execution_mode_begin(); + while (mode) { + if (mode->opcode() != spv::Op::OpExecutionMode && + mode->opcode() != spv::Op::OpExecutionModeId) { + break; + } + if (spv::ExecutionMode(mode->GetSingleWordInOperand(1)) == + spv::ExecutionMode::MaximallyReconvergesKHR) { + mode = context()->KillInst(mode); + changed = true; + } else { + mode = mode->NextNode(); + } + } + + changed |= + context()->RemoveExtension(Extension::kSPV_KHR_maximal_reconvergence); + return changed; +} +} // namespace opt +} // namespace spvtools diff --git a/3rdparty/spirv-tools/source/opt/modify_maximal_reconvergence.h b/3rdparty/spirv-tools/source/opt/modify_maximal_reconvergence.h new file mode 100644 index 000000000..8d9a698e9 --- /dev/null +++ b/3rdparty/spirv-tools/source/opt/modify_maximal_reconvergence.h @@ -0,0 +1,53 @@ +// Copyright (c) 2024 Google LLC +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef LIBSPIRV_OPT_MODIFY_MAXIMAL_RECONVERGENCE_H_ +#define LIBSPIRV_OPT_MODIFY_MAXIMAL_RECONVERGENCE_H_ + +#include "pass.h" + +namespace spvtools { +namespace opt { + +// Modifies entry points to either add or remove MaximallyReconvergesKHR +// +// This pass will either add or remove MaximallyReconvergesKHR to all entry +// points in the module. When adding the execution mode, it does not attempt to +// determine whether any ray tracing invocation repack instructions might be +// executed because it is a runtime restriction. That is left to the user. +class ModifyMaximalReconvergence : public Pass { + public: + const char* name() const override { return "modify-maximal-reconvergence"; } + Status Process() override; + + explicit ModifyMaximalReconvergence(bool add = true) : Pass(), add_(add) {} + + IRContext::Analysis GetPreservedAnalyses() override { + return IRContext::kAnalysisDefUse | + IRContext::kAnalysisInstrToBlockMapping | + IRContext::kAnalysisDecorations | IRContext::kAnalysisCombinators | + IRContext::kAnalysisCFG | IRContext::kAnalysisNameMap | + IRContext::kAnalysisConstants | IRContext::kAnalysisTypes; + } + + private: + bool AddMaximalReconvergence(); + bool RemoveMaximalReconvergence(); + + bool add_; +}; +} // namespace opt +} // namespace spvtools + +#endif // LIBSPIRV_OPT_MODIFY_MAXIMAL_RECONVERGENCE_H_ diff --git a/3rdparty/spirv-tools/source/opt/optimizer.cpp b/3rdparty/spirv-tools/source/opt/optimizer.cpp index 7cfc89f75..429a6cba8 100644 --- a/3rdparty/spirv-tools/source/opt/optimizer.cpp +++ b/3rdparty/spirv-tools/source/opt/optimizer.cpp @@ -606,6 +606,25 @@ bool Optimizer::RegisterPassFromFlag(const std::string& flag, return false; } RegisterPass(CreateSwitchDescriptorSetPass(from_set, to_set)); + } else if (pass_name == "modify-maximal-reconvergence") { + if (pass_args.size() == 0) { + Error(consumer(), nullptr, {}, + "--modify-maximal-reconvergence requires an argument"); + return false; + } + if (pass_args == "add") { + RegisterPass(CreateModifyMaximalReconvergencePass(true)); + } else if (pass_args == "remove") { + RegisterPass(CreateModifyMaximalReconvergencePass(false)); + } else { + Errorf(consumer(), nullptr, {}, + "Invalid argument for --modify-maximal-reconvergence: %s (must be " + "'add' or 'remove')", + pass_args.c_str()); + return false; + } + } else if (pass_name == "trim-capabilities") { + RegisterPass(CreateTrimCapabilitiesPass()); } else { Errorf(consumer(), nullptr, {}, "Unknown flag '--%s'. Use --help for a list of valid flags", @@ -1141,6 +1160,11 @@ Optimizer::PassToken CreateInvocationInterlockPlacementPass() { return MakeUnique( MakeUnique()); } + +Optimizer::PassToken CreateModifyMaximalReconvergencePass(bool add) { + return MakeUnique( + MakeUnique(add)); +} } // namespace spvtools extern "C" { diff --git a/3rdparty/spirv-tools/source/opt/passes.h b/3rdparty/spirv-tools/source/opt/passes.h index 305f57827..9d027fbf4 100644 --- a/3rdparty/spirv-tools/source/opt/passes.h +++ b/3rdparty/spirv-tools/source/opt/passes.h @@ -65,6 +65,7 @@ #include "source/opt/loop_unroller.h" #include "source/opt/loop_unswitch_pass.h" #include "source/opt/merge_return_pass.h" +#include "source/opt/modify_maximal_reconvergence.h" #include "source/opt/null_pass.h" #include "source/opt/private_to_local_pass.h" #include "source/opt/reduce_load_size.h" diff --git a/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.cpp b/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.cpp index 19f8569e0..24f9e4670 100644 --- a/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.cpp +++ b/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.cpp @@ -448,6 +448,17 @@ void TrimCapabilitiesPass::addInstructionRequirementsForOperand( return; } + // If the Vulkan memory model is declared and any instruction uses Device + // scope, the VulkanMemoryModelDeviceScope capability must be declared. This + // rule cannot be covered by the grammar, so must be checked explicitly. + if (operand.type == SPV_OPERAND_TYPE_SCOPE_ID) { + const Instruction* memory_model = context()->GetMemoryModel(); + if (memory_model && memory_model->GetSingleWordInOperand(1u) == + uint32_t(spv::MemoryModel::Vulkan)) { + capabilities->insert(spv::Capability::VulkanMemoryModelDeviceScope); + } + } + // case 1: Operand is a single value, can directly lookup. if (!spvOperandIsConcreteMask(operand.type)) { const spv_operand_desc_t* desc = {}; diff --git a/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.h b/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.h index 9f2373299..3a8460f3b 100644 --- a/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.h +++ b/3rdparty/spirv-tools/source/opt/trim_capabilities_pass.h @@ -97,7 +97,8 @@ class TrimCapabilitiesPass : public Pass { spv::Capability::StorageInputOutput16, spv::Capability::StoragePushConstant16, spv::Capability::StorageUniform16, - spv::Capability::StorageUniformBufferBlock16 + spv::Capability::StorageUniformBufferBlock16, + spv::Capability::VulkanMemoryModelDeviceScope // clang-format on }; diff --git a/3rdparty/spirv-tools/source/val/validate.cpp b/3rdparty/spirv-tools/source/val/validate.cpp index a5f320b9f..ccf26fbf1 100644 --- a/3rdparty/spirv-tools/source/val/validate.cpp +++ b/3rdparty/spirv-tools/source/val/validate.cpp @@ -141,6 +141,10 @@ spv_result_t ValidateEntryPoints(ValidationState_t& _) { } } + if (auto error = ValidateFloatControls2(_)) { + return error; + } + return SPV_SUCCESS; } diff --git a/3rdparty/spirv-tools/source/val/validate.h b/3rdparty/spirv-tools/source/val/validate.h index 6b7d7cdac..52267c8ab 100644 --- a/3rdparty/spirv-tools/source/val/validate.h +++ b/3rdparty/spirv-tools/source/val/validate.h @@ -82,6 +82,18 @@ spv_result_t ValidateAdjacency(ValidationState_t& _); /// @return SPV_SUCCESS if no errors are found. spv_result_t ValidateInterfaces(ValidationState_t& _); +/// @brief Validates entry point call tree requirements of +/// SPV_KHR_float_controls2 +/// +/// Checks that no entry point using FPFastMathDefault uses: +/// * FPFastMathMode Fast +/// * NoContraction +/// +/// @param[in] _ the validation state of the module +/// +/// @return SPV_SUCCESS if no errors are found. +spv_result_t ValidateFloatControls2(ValidationState_t& _); + /// @brief Validates memory instructions /// /// @param[in] _ the validation state of the module diff --git a/3rdparty/spirv-tools/source/val/validate_annotation.cpp b/3rdparty/spirv-tools/source/val/validate_annotation.cpp index 73d0285a1..106004d06 100644 --- a/3rdparty/spirv-tools/source/val/validate_annotation.cpp +++ b/3rdparty/spirv-tools/source/val/validate_annotation.cpp @@ -267,6 +267,34 @@ spv_result_t ValidateDecorate(ValidationState_t& _, const Instruction* inst) { } } + if (decoration == spv::Decoration::FPFastMathMode) { + if (_.HasDecoration(target_id, spv::Decoration::NoContraction)) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "FPFastMathMode and NoContraction cannot decorate the same " + "target"; + } + auto mask = inst->GetOperandAs(2); + if ((mask & spv::FPFastMathModeMask::AllowTransform) != + spv::FPFastMathModeMask::MaskNone && + ((mask & (spv::FPFastMathModeMask::AllowContract | + spv::FPFastMathModeMask::AllowReassoc)) != + (spv::FPFastMathModeMask::AllowContract | + spv::FPFastMathModeMask::AllowReassoc))) { + return _.diag(SPV_ERROR_INVALID_DATA, inst) + << "AllowReassoc and AllowContract must be specified when " + "AllowTransform is specified"; + } + } + + // This is checked from both sides since we register decorations as we go. + if (decoration == spv::Decoration::NoContraction) { + if (_.HasDecoration(target_id, spv::Decoration::FPFastMathMode)) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "FPFastMathMode and NoContraction cannot decorate the same " + "target"; + } + } + if (DecorationTakesIdParameters(decoration)) { return _.diag(SPV_ERROR_INVALID_ID, inst) << "Decorations taking ID parameters may not be used with " diff --git a/3rdparty/spirv-tools/source/val/validate_builtins.cpp b/3rdparty/spirv-tools/source/val/validate_builtins.cpp index 3e8171250..42fbc52a7 100644 --- a/3rdparty/spirv-tools/source/val/validate_builtins.cpp +++ b/3rdparty/spirv-tools/source/val/validate_builtins.cpp @@ -118,13 +118,15 @@ typedef enum VUIDError_ { VUIDErrorMax, } VUIDError; -const static uint32_t NumVUIDBuiltins = 36; +const static uint32_t NumVUIDBuiltins = 39; typedef struct { spv::BuiltIn builtIn; uint32_t vuid[VUIDErrorMax]; // execution mode, storage class, type VUIDs } BuiltinVUIDMapping; +// Many built-ins have the same checks (Storage Class, Type, etc) +// This table provides a nice LUT for the VUIDs std::array builtinVUIDInfo = {{ // clang-format off {spv::BuiltIn::SubgroupEqMask, {0, 4370, 4371}}, @@ -163,8 +165,11 @@ std::array builtinVUIDInfo = {{ {spv::BuiltIn::CullMaskKHR, {6735, 6736, 6737}}, {spv::BuiltIn::BaryCoordKHR, {4154, 4155, 4156}}, {spv::BuiltIn::BaryCoordNoPerspKHR, {4160, 4161, 4162}}, - // clang-format off -} }; + {spv::BuiltIn::PrimitivePointIndicesEXT, {7041, 7043, 7044}}, + {spv::BuiltIn::PrimitiveLineIndicesEXT, {7047, 7049, 7050}}, + {spv::BuiltIn::PrimitiveTriangleIndicesEXT, {7053, 7055, 7056}}, + // clang-format on +}}; uint32_t GetVUIDForBuiltin(spv::BuiltIn builtIn, VUIDError type) { uint32_t vuid = 0; @@ -356,6 +361,9 @@ class BuiltInsValidator { spv_result_t ValidateRayTracingBuiltinsAtDefinition( const Decoration& decoration, const Instruction& inst); + spv_result_t ValidateMeshShadingEXTBuiltinsAtDefinition( + const Decoration& decoration, const Instruction& inst); + // The following section contains functions which are called when id defined // by |referenced_inst| is // 1. referenced by |referenced_from_inst| @@ -546,6 +554,11 @@ class BuiltInsValidator { const Instruction& referenced_inst, const Instruction& referenced_from_inst); + spv_result_t ValidateMeshShadingEXTBuiltinsAtReference( + const Decoration& decoration, const Instruction& built_in_inst, + const Instruction& referenced_inst, + const Instruction& referenced_from_inst); + // Validates that |built_in_inst| is not (even indirectly) referenced from // within a function which can be called with |execution_model|. // @@ -581,6 +594,10 @@ class BuiltInsValidator { spv_result_t ValidateI32Arr( const Decoration& decoration, const Instruction& inst, const std::function& diag); + spv_result_t ValidateArrayedI32Vec( + const Decoration& decoration, const Instruction& inst, + uint32_t num_components, + const std::function& diag); spv_result_t ValidateOptionalArrayedI32( const Decoration& decoration, const Instruction& inst, const std::function& diag); @@ -909,6 +926,45 @@ spv_result_t BuiltInsValidator::ValidateI32Vec( return SPV_SUCCESS; } +spv_result_t BuiltInsValidator::ValidateArrayedI32Vec( + const Decoration& decoration, const Instruction& inst, + uint32_t num_components, + const std::function& diag) { + uint32_t underlying_type = 0; + if (spv_result_t error = + GetUnderlyingType(_, decoration, inst, &underlying_type)) { + return error; + } + + const Instruction* const type_inst = _.FindDef(underlying_type); + if (type_inst->opcode() != spv::Op::OpTypeArray) { + return diag(GetDefinitionDesc(decoration, inst) + " is not an array."); + } + + const uint32_t component_type = type_inst->word(2); + if (!_.IsIntVectorType(component_type)) { + return diag(GetDefinitionDesc(decoration, inst) + " is not an int vector."); + } + + const uint32_t actual_num_components = _.GetDimension(component_type); + if (_.GetDimension(component_type) != num_components) { + std::ostringstream ss; + ss << GetDefinitionDesc(decoration, inst) << " has " + << actual_num_components << " components."; + return diag(ss.str()); + } + + const uint32_t bit_width = _.GetBitWidth(component_type); + if (bit_width != 32) { + std::ostringstream ss; + ss << GetDefinitionDesc(decoration, inst) + << " has components with bit width " << bit_width << "."; + return diag(ss.str()); + } + + return SPV_SUCCESS; +} + spv_result_t BuiltInsValidator::ValidateOptionalArrayedF32Vec( const Decoration& decoration, const Instruction& inst, uint32_t num_components, @@ -4108,6 +4164,119 @@ spv_result_t BuiltInsValidator::ValidateRayTracingBuiltinsAtReference( return SPV_SUCCESS; } +spv_result_t BuiltInsValidator::ValidateMeshShadingEXTBuiltinsAtDefinition( + const Decoration& decoration, const Instruction& inst) { + if (spvIsVulkanEnv(_.context()->target_env)) { + const spv::BuiltIn builtin = spv::BuiltIn(decoration.params()[0]); + uint32_t vuid = GetVUIDForBuiltin(builtin, VUIDErrorType); + if (builtin == spv::BuiltIn::PrimitivePointIndicesEXT) { + if (spv_result_t error = ValidateI32Arr( + decoration, inst, + [this, &inst, &decoration, + &vuid](const std::string& message) -> spv_result_t { + return _.diag(SPV_ERROR_INVALID_DATA, &inst) + << _.VkErrorID(vuid) << "According to the " + << spvLogStringForEnv(_.context()->target_env) + << " spec BuiltIn " + << _.grammar().lookupOperandName( + SPV_OPERAND_TYPE_BUILT_IN, decoration.params()[0]) + << " variable needs to be a 32-bit int array." + << message; + })) { + return error; + } + } + if (builtin == spv::BuiltIn::PrimitiveLineIndicesEXT) { + if (spv_result_t error = ValidateArrayedI32Vec( + decoration, inst, 2, + [this, &inst, &decoration, + &vuid](const std::string& message) -> spv_result_t { + return _.diag(SPV_ERROR_INVALID_DATA, &inst) + << _.VkErrorID(vuid) << "According to the " + << spvLogStringForEnv(_.context()->target_env) + << " spec BuiltIn " + << _.grammar().lookupOperandName( + SPV_OPERAND_TYPE_BUILT_IN, decoration.params()[0]) + << " variable needs to be a 2-component 32-bit int " + "array." + << message; + })) { + return error; + } + } + if (builtin == spv::BuiltIn::PrimitiveTriangleIndicesEXT) { + if (spv_result_t error = ValidateArrayedI32Vec( + decoration, inst, 3, + [this, &inst, &decoration, + &vuid](const std::string& message) -> spv_result_t { + return _.diag(SPV_ERROR_INVALID_DATA, &inst) + << _.VkErrorID(vuid) << "According to the " + << spvLogStringForEnv(_.context()->target_env) + << " spec BuiltIn " + << _.grammar().lookupOperandName( + SPV_OPERAND_TYPE_BUILT_IN, decoration.params()[0]) + << " variable needs to be a 3-component 32-bit int " + "array." + << message; + })) { + return error; + } + } + } + // Seed at reference checks with this built-in. + return ValidateMeshShadingEXTBuiltinsAtReference(decoration, inst, inst, + inst); +} + +spv_result_t BuiltInsValidator::ValidateMeshShadingEXTBuiltinsAtReference( + const Decoration& decoration, const Instruction& built_in_inst, + const Instruction& referenced_inst, + const Instruction& referenced_from_inst) { + if (spvIsVulkanEnv(_.context()->target_env)) { + const spv::BuiltIn builtin = spv::BuiltIn(decoration.params()[0]); + const spv::StorageClass storage_class = + GetStorageClass(referenced_from_inst); + if (storage_class != spv::StorageClass::Max && + storage_class != spv::StorageClass::Output) { + uint32_t vuid = GetVUIDForBuiltin(builtin, VUIDErrorStorageClass); + return _.diag(SPV_ERROR_INVALID_DATA, &referenced_from_inst) + << _.VkErrorID(vuid) << spvLogStringForEnv(_.context()->target_env) + << " spec allows BuiltIn " + << _.grammar().lookupOperandName(SPV_OPERAND_TYPE_BUILT_IN, + uint32_t(builtin)) + << " to be only used for variables with Output storage class. " + << GetReferenceDesc(decoration, built_in_inst, referenced_inst, + referenced_from_inst) + << " " << GetStorageClassDesc(referenced_from_inst); + } + + for (const spv::ExecutionModel execution_model : execution_models_) { + if (execution_model != spv::ExecutionModel::MeshEXT) { + uint32_t vuid = GetVUIDForBuiltin(builtin, VUIDErrorExecutionModel); + return _.diag(SPV_ERROR_INVALID_DATA, &referenced_from_inst) + << _.VkErrorID(vuid) + << spvLogStringForEnv(_.context()->target_env) + << " spec allows BuiltIn " + << _.grammar().lookupOperandName(SPV_OPERAND_TYPE_BUILT_IN, + uint32_t(builtin)) + << " to be used only with MeshEXT execution model. " + << GetReferenceDesc(decoration, built_in_inst, referenced_inst, + referenced_from_inst, execution_model); + } + } + } + + if (function_id_ == 0) { + // Propagate this rule to all dependant ids in the global scope. + id_to_at_reference_checks_[referenced_from_inst.id()].push_back( + std::bind(&BuiltInsValidator::ValidateMeshShadingEXTBuiltinsAtReference, + this, decoration, built_in_inst, referenced_from_inst, + std::placeholders::_1)); + } + + return SPV_SUCCESS; +} + spv_result_t BuiltInsValidator::ValidateSingleBuiltInAtDefinition( const Decoration& decoration, const Instruction& inst) { const spv::BuiltIn label = spv::BuiltIn(decoration.params()[0]); @@ -4283,6 +4452,11 @@ spv_result_t BuiltInsValidator::ValidateSingleBuiltInAtDefinition( case spv::BuiltIn::CullMaskKHR: { return ValidateRayTracingBuiltinsAtDefinition(decoration, inst); } + case spv::BuiltIn::PrimitivePointIndicesEXT: + case spv::BuiltIn::PrimitiveLineIndicesEXT: + case spv::BuiltIn::PrimitiveTriangleIndicesEXT: { + return ValidateMeshShadingEXTBuiltinsAtDefinition(decoration, inst); + } case spv::BuiltIn::PrimitiveShadingRateKHR: { return ValidatePrimitiveShadingRateAtDefinition(decoration, inst); } diff --git a/3rdparty/spirv-tools/source/val/validate_cfg.cpp b/3rdparty/spirv-tools/source/val/validate_cfg.cpp index 7b3f7480f..9b7161fc4 100644 --- a/3rdparty/spirv-tools/source/val/validate_cfg.cpp +++ b/3rdparty/spirv-tools/source/val/validate_cfg.cpp @@ -190,6 +190,8 @@ spv_result_t ValidateBranchConditional(ValidationState_t& _, "ID of an OpLabel instruction"; } + // A similar requirement for SPV_KHR_maximal_reconvergence is deferred until + // entry point call trees have been reconrded. if (_.version() >= SPV_SPIRV_VERSION_WORD(1, 6) && true_id == false_id) { return _.diag(SPV_ERROR_INVALID_ID, inst) << "In SPIR-V 1.6 or later, True Label and False Label must be " @@ -875,6 +877,95 @@ spv_result_t StructuredControlFlowChecks( return SPV_SUCCESS; } +spv_result_t MaximalReconvergenceChecks(ValidationState_t& _) { + // Find all the entry points with the MaximallyReconvergencesKHR execution + // mode. + std::unordered_set maximal_funcs; + std::unordered_set maximal_entry_points; + for (auto entry_point : _.entry_points()) { + const auto* exec_modes = _.GetExecutionModes(entry_point); + if (exec_modes && + exec_modes->count(spv::ExecutionMode::MaximallyReconvergesKHR)) { + maximal_entry_points.insert(entry_point); + maximal_funcs.insert(entry_point); + } + } + + if (maximal_entry_points.empty()) { + return SPV_SUCCESS; + } + + // Find all the functions reachable from a maximal reconvergence entry point. + for (const auto& func : _.functions()) { + const auto& entry_points = _.EntryPointReferences(func.id()); + for (auto id : entry_points) { + if (maximal_entry_points.count(id)) { + maximal_funcs.insert(func.id()); + break; + } + } + } + + // Check for conditional branches with the same true and false targets. + for (const auto& inst : _.ordered_instructions()) { + if (inst.opcode() == spv::Op::OpBranchConditional) { + const auto true_id = inst.GetOperandAs(1); + const auto false_id = inst.GetOperandAs(2); + if (true_id == false_id && maximal_funcs.count(inst.function()->id())) { + return _.diag(SPV_ERROR_INVALID_ID, &inst) + << "In entry points using the MaximallyReconvergesKHR execution " + "mode, True Label and False Label must be different labels"; + } + } + } + + // Check for invalid multiple predecessors. Only loop headers, continue + // targets, merge targets or switch targets or defaults may have multiple + // unique predecessors. + for (const auto& func : _.functions()) { + if (!maximal_funcs.count(func.id())) continue; + + for (const auto* block : func.ordered_blocks()) { + std::unordered_set unique_preds; + const auto* preds = block->predecessors(); + if (!preds) continue; + + for (const auto* pred : *preds) { + unique_preds.insert(pred->id()); + } + if (unique_preds.size() < 2) continue; + + const auto* terminator = block->terminator(); + const auto index = terminator - &_.ordered_instructions()[0]; + const auto* pre_terminator = &_.ordered_instructions()[index - 1]; + if (pre_terminator->opcode() == spv::Op::OpLoopMerge) continue; + + const auto* label = _.FindDef(block->id()); + bool ok = false; + for (const auto& pair : label->uses()) { + const auto* use_inst = pair.first; + switch (use_inst->opcode()) { + case spv::Op::OpSelectionMerge: + case spv::Op::OpLoopMerge: + case spv::Op::OpSwitch: + ok = true; + break; + default: + break; + } + } + if (!ok) { + return _.diag(SPV_ERROR_INVALID_CFG, label) + << "In entry points using the MaximallyReconvergesKHR " + "execution mode, this basic block must not have multiple " + "unique predecessors"; + } + } + } + + return SPV_SUCCESS; +} + spv_result_t PerformCfgChecks(ValidationState_t& _) { for (auto& function : _.functions()) { // Check all referenced blocks are defined within a function @@ -999,6 +1090,11 @@ spv_result_t PerformCfgChecks(ValidationState_t& _) { return error; } } + + if (auto error = MaximalReconvergenceChecks(_)) { + return error; + } + return SPV_SUCCESS; } diff --git a/3rdparty/spirv-tools/source/val/validate_instruction.cpp b/3rdparty/spirv-tools/source/val/validate_instruction.cpp index 8710ffa44..5bc4d2cef 100644 --- a/3rdparty/spirv-tools/source/val/validate_instruction.cpp +++ b/3rdparty/spirv-tools/source/val/validate_instruction.cpp @@ -470,7 +470,8 @@ spv_result_t InstructionPass(ValidationState_t& _, const Instruction* inst) { } _.set_addressing_model(inst->GetOperandAs(0)); _.set_memory_model(inst->GetOperandAs(1)); - } else if (opcode == spv::Op::OpExecutionMode) { + } else if (opcode == spv::Op::OpExecutionMode || + opcode == spv::Op::OpExecutionModeId) { const uint32_t entry_point = inst->word(1); _.RegisterExecutionModeForEntryPoint(entry_point, spv::ExecutionMode(inst->word(2))); diff --git a/3rdparty/spirv-tools/source/val/validate_memory_semantics.cpp b/3rdparty/spirv-tools/source/val/validate_memory_semantics.cpp index c4f22a611..dab7b5a19 100644 --- a/3rdparty/spirv-tools/source/val/validate_memory_semantics.cpp +++ b/3rdparty/spirv-tools/source/val/validate_memory_semantics.cpp @@ -203,15 +203,12 @@ spv_result_t ValidateMemorySemantics(ValidationState_t& _, "storage class"; } -#if 0 - // TODO(atgoo@github.com): this check fails Vulkan CTS, reenable once fixed. if (opcode == spv::Op::OpControlBarrier && value && !includes_storage_class) { return _.diag(SPV_ERROR_INVALID_DATA, inst) - << spvOpcodeString(opcode) + << _.VkErrorID(4650) << spvOpcodeString(opcode) << ": expected Memory Semantics to include a Vulkan-supported " "storage class if Memory Semantics is not None"; } -#endif } if (opcode == spv::Op::OpAtomicFlagClear && diff --git a/3rdparty/spirv-tools/source/val/validate_mode_setting.cpp b/3rdparty/spirv-tools/source/val/validate_mode_setting.cpp index d757d4f82..82c6c3f0e 100644 --- a/3rdparty/spirv-tools/source/val/validate_mode_setting.cpp +++ b/3rdparty/spirv-tools/source/val/validate_mode_setting.cpp @@ -340,29 +340,92 @@ spv_result_t ValidateExecutionMode(ValidationState_t& _, const auto mode = inst->GetOperandAs(1); if (inst->opcode() == spv::Op::OpExecutionModeId) { + bool valid_mode = false; + switch (mode) { + case spv::ExecutionMode::SubgroupsPerWorkgroupId: + case spv::ExecutionMode::LocalSizeHintId: + case spv::ExecutionMode::LocalSizeId: + case spv::ExecutionMode::FPFastMathDefault: + valid_mode = true; + break; + default: + valid_mode = false; + break; + } + if (!valid_mode) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "OpExecutionModeId is only valid when the Mode operand is an " + "execution mode that takes Extra Operands that are id " + "operands."; + } + size_t operand_count = inst->operands().size(); for (size_t i = 2; i < operand_count; ++i) { - const auto operand_id = inst->GetOperandAs(2); + const auto operand_id = inst->GetOperandAs(i); const auto* operand_inst = _.FindDef(operand_id); - if (mode == spv::ExecutionMode::SubgroupsPerWorkgroupId || - mode == spv::ExecutionMode::LocalSizeHintId || - mode == spv::ExecutionMode::LocalSizeId) { - if (!spvOpcodeIsConstant(operand_inst->opcode())) { - return _.diag(SPV_ERROR_INVALID_ID, inst) - << "For OpExecutionModeId all Extra Operand ids must be " - "constant " - "instructions."; - } - } else { - return _.diag(SPV_ERROR_INVALID_ID, inst) - << "OpExecutionModeId is only valid when the Mode operand is an " - "execution mode that takes Extra Operands that are id " - "operands."; + switch (mode) { + case spv::ExecutionMode::SubgroupsPerWorkgroupId: + case spv::ExecutionMode::LocalSizeHintId: + case spv::ExecutionMode::LocalSizeId: + if (!spvOpcodeIsConstant(operand_inst->opcode())) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "For OpExecutionModeId all Extra Operand ids must be " + "constant instructions."; + } + break; + case spv::ExecutionMode::FPFastMathDefault: + if (i == 2) { + if (!_.IsFloatScalarType(operand_id)) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "The Target Type operand must be a floating-point " + "scalar type"; + } + } else { + bool is_int32 = false; + bool is_const = false; + uint32_t value = 0; + std::tie(is_int32, is_const, value) = + _.EvalInt32IfConst(operand_id); + if (is_int32 && is_const) { + // Valid values include up to 0x00040000 (AllowTransform). + uint32_t invalid_mask = 0xfff80000; + if ((invalid_mask & value) != 0) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "The Fast Math Default operand is an invalid bitmask " + "value"; + } + if (value & + static_cast(spv::FPFastMathModeMask::Fast)) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "The Fast Math Default operand must not include Fast"; + } + const auto reassoc_contract = + spv::FPFastMathModeMask::AllowContract | + spv::FPFastMathModeMask::AllowReassoc; + if ((value & static_cast( + spv::FPFastMathModeMask::AllowTransform)) != 0 && + ((value & static_cast(reassoc_contract)) != + static_cast(reassoc_contract))) { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "The Fast Math Default operand must include " + "AllowContract and AllowReassoc when AllowTransform " + "is specified"; + } + } else { + return _.diag(SPV_ERROR_INVALID_ID, inst) + << "The Fast Math Default operand must be a " + "non-specialization constant"; + } + } + break; + default: + break; } } } else if (mode == spv::ExecutionMode::SubgroupsPerWorkgroupId || mode == spv::ExecutionMode::LocalSizeHintId || - mode == spv::ExecutionMode::LocalSizeId) { + mode == spv::ExecutionMode::LocalSizeId || + mode == spv::ExecutionMode::FPFastMathDefault) { return _.diag(SPV_ERROR_INVALID_DATA, inst) << "OpExecutionMode is only valid when the Mode operand is an " "execution mode that takes no Extra Operands, or takes Extra " @@ -494,6 +557,17 @@ spv_result_t ValidateExecutionMode(ValidationState_t& _, "model."; } break; + case spv::ExecutionMode::QuadDerivativesKHR: + if (!std::all_of(models->begin(), models->end(), + [](const spv::ExecutionModel& model) { + return (model == spv::ExecutionModel::Fragment || + model == spv::ExecutionModel::GLCompute); + })) { + return _.diag(SPV_ERROR_INVALID_DATA, inst) + << "Execution mode can only be used with the Fragment or " + "GLCompute execution model."; + } + break; case spv::ExecutionMode::PixelCenterInteger: case spv::ExecutionMode::OriginUpperLeft: case spv::ExecutionMode::OriginLowerLeft: @@ -518,6 +592,7 @@ spv_result_t ValidateExecutionMode(ValidationState_t& _, case spv::ExecutionMode::StencilRefUnchangedBackAMD: case spv::ExecutionMode::StencilRefGreaterBackAMD: case spv::ExecutionMode::StencilRefLessBackAMD: + case spv::ExecutionMode::RequireFullQuadsKHR: if (!std::all_of(models->begin(), models->end(), [](const spv::ExecutionModel& model) { return model == spv::ExecutionModel::Fragment; @@ -579,6 +654,20 @@ spv_result_t ValidateExecutionMode(ValidationState_t& _, break; } + if (mode == spv::ExecutionMode::FPFastMathDefault) { + const auto* modes = _.GetExecutionModes(entry_point_id); + if (modes && modes->count(spv::ExecutionMode::ContractionOff)) { + return _.diag(SPV_ERROR_INVALID_DATA, inst) + << "FPFastMathDefault and ContractionOff execution modes cannot " + "be applied to the same entry point"; + } + if (modes && modes->count(spv::ExecutionMode::SignedZeroInfNanPreserve)) { + return _.diag(SPV_ERROR_INVALID_DATA, inst) + << "FPFastMathDefault and SignedZeroInfNanPreserve execution " + "modes cannot be applied to the same entry point"; + } + } + if (spvIsVulkanEnv(_.context()->target_env)) { if (mode == spv::ExecutionMode::OriginLowerLeft) { return _.diag(SPV_ERROR_INVALID_DATA, inst) @@ -636,6 +725,70 @@ spv_result_t ValidateMemoryModel(ValidationState_t& _, } // namespace +spv_result_t ValidateFloatControls2(ValidationState_t& _) { + std::unordered_set fp_fast_math_default_entry_points; + for (auto entry_point : _.entry_points()) { + const auto* exec_modes = _.GetExecutionModes(entry_point); + if (exec_modes && + exec_modes->count(spv::ExecutionMode::FPFastMathDefault)) { + fp_fast_math_default_entry_points.insert(entry_point); + } + } + + std::vector> worklist; + for (const auto& inst : _.ordered_instructions()) { + if (inst.opcode() != spv::Op::OpDecorate) { + continue; + } + + const auto decoration = inst.GetOperandAs(1); + const auto target_id = inst.GetOperandAs(0); + const auto target = _.FindDef(target_id); + if (decoration == spv::Decoration::NoContraction) { + worklist.push_back(std::make_pair(target, decoration)); + } else if (decoration == spv::Decoration::FPFastMathMode) { + auto mask = inst.GetOperandAs(2); + if ((mask & spv::FPFastMathModeMask::Fast) != + spv::FPFastMathModeMask::MaskNone) { + worklist.push_back(std::make_pair(target, decoration)); + } + } + } + + std::unordered_set visited; + while (!worklist.empty()) { + const auto inst = worklist.back().first; + const auto decoration = worklist.back().second; + worklist.pop_back(); + + if (!visited.insert(inst).second) { + continue; + } + + const auto function = inst->function(); + if (function) { + const auto& entry_points = _.FunctionEntryPoints(function->id()); + for (auto entry_point : entry_points) { + if (fp_fast_math_default_entry_points.count(entry_point)) { + const std::string dec = decoration == spv::Decoration::NoContraction + ? "NoContraction" + : "FPFastMathMode Fast"; + return _.diag(SPV_ERROR_INVALID_DATA, inst) + << dec + << " cannot be used by an entry point with the " + "FPFastMathDefault execution mode"; + } + } + } else { + for (const auto& pair : inst->uses()) { + worklist.push_back(std::make_pair(pair.first, decoration)); + } + } + } + + return SPV_SUCCESS; +} + spv_result_t ModeSettingPass(ValidationState_t& _, const Instruction* inst) { switch (inst->opcode()) { case spv::Op::OpEntryPoint: diff --git a/3rdparty/spirv-tools/source/val/validate_non_uniform.cpp b/3rdparty/spirv-tools/source/val/validate_non_uniform.cpp index 2c36ce332..74449e9dc 100644 --- a/3rdparty/spirv-tools/source/val/validate_non_uniform.cpp +++ b/3rdparty/spirv-tools/source/val/validate_non_uniform.cpp @@ -422,9 +422,14 @@ spv_result_t NonUniformPass(ValidationState_t& _, const Instruction* inst) { const spv::Op opcode = inst->opcode(); if (spvOpcodeIsNonUniformGroupOperation(opcode)) { - const uint32_t execution_scope = inst->GetOperandAs(2); - if (auto error = ValidateExecutionScope(_, inst, execution_scope)) { - return error; + // OpGroupNonUniformQuadAllKHR and OpGroupNonUniformQuadAnyKHR don't have + // scope paramter + if ((opcode != spv::Op::OpGroupNonUniformQuadAllKHR) && + (opcode != spv::Op::OpGroupNonUniformQuadAnyKHR)) { + const uint32_t execution_scope = inst->GetOperandAs(2); + if (auto error = ValidateExecutionScope(_, inst, execution_scope)) { + return error; + } } } diff --git a/3rdparty/spirv-tools/source/val/validate_scopes.cpp b/3rdparty/spirv-tools/source/val/validate_scopes.cpp index 40c49d1ff..6b493538a 100644 --- a/3rdparty/spirv-tools/source/val/validate_scopes.cpp +++ b/3rdparty/spirv-tools/source/val/validate_scopes.cpp @@ -97,8 +97,10 @@ spv_result_t ValidateExecutionScope(ValidationState_t& _, // Vulkan 1.1 specific rules if (_.context()->target_env != SPV_ENV_VULKAN_1_0) { // Scope for Non Uniform Group Operations must be limited to Subgroup - if (spvOpcodeIsNonUniformGroupOperation(opcode) && - value != spv::Scope::Subgroup) { + if ((spvOpcodeIsNonUniformGroupOperation(opcode) && + (opcode != spv::Op::OpGroupNonUniformQuadAllKHR) && + (opcode != spv::Op::OpGroupNonUniformQuadAnyKHR)) && + (value != spv::Scope::Subgroup)) { return _.diag(SPV_ERROR_INVALID_DATA, inst) << _.VkErrorID(4642) << spvOpcodeString(opcode) << ": in Vulkan environment Execution scope is limited to " @@ -178,6 +180,8 @@ spv_result_t ValidateExecutionScope(ValidationState_t& _, // Scope for execution must be limited to Workgroup or Subgroup for // non-uniform operations if (spvOpcodeIsNonUniformGroupOperation(opcode) && + opcode != spv::Op::OpGroupNonUniformQuadAllKHR && + opcode != spv::Op::OpGroupNonUniformQuadAnyKHR && value != spv::Scope::Subgroup && value != spv::Scope::Workgroup) { return _.diag(SPV_ERROR_INVALID_DATA, inst) << spvOpcodeString(opcode) diff --git a/3rdparty/spirv-tools/source/val/validation_state.cpp b/3rdparty/spirv-tools/source/val/validation_state.cpp index 6685985b6..852860e17 100644 --- a/3rdparty/spirv-tools/source/val/validation_state.cpp +++ b/3rdparty/spirv-tools/source/val/validation_state.cpp @@ -2125,6 +2125,8 @@ std::string ValidationState_t::VkErrorID(uint32_t id, return VUID_WRAP(VUID-StandaloneSpirv-None-04644); case 4645: return VUID_WRAP(VUID-StandaloneSpirv-None-04645); + case 4650: + return VUID_WRAP(VUID-StandaloneSpirv-OpControlBarrier-04650); case 4651: return VUID_WRAP(VUID-StandaloneSpirv-OpVariable-04651); case 4652: @@ -2257,6 +2259,24 @@ std::string ValidationState_t::VkErrorID(uint32_t id, return VUID_WRAP(VUID-StandaloneSpirv-PushConstant-06808); case 6925: return VUID_WRAP(VUID-StandaloneSpirv-Uniform-06925); + case 7041: + return VUID_WRAP(VUID-PrimitivePointIndicesEXT-PrimitivePointIndicesEXT-07041); + case 7043: + return VUID_WRAP(VUID-PrimitivePointIndicesEXT-PrimitivePointIndicesEXT-07043); + case 7044: + return VUID_WRAP(VUID-PrimitivePointIndicesEXT-PrimitivePointIndicesEXT-07044); + case 7047: + return VUID_WRAP(VUID-PrimitiveLineIndicesEXT-PrimitiveLineIndicesEXT-07047); + case 7049: + return VUID_WRAP(VUID-PrimitiveLineIndicesEXT-PrimitiveLineIndicesEXT-07049); + case 7050: + return VUID_WRAP(VUID-PrimitiveLineIndicesEXT-PrimitiveLineIndicesEXT-07050); + case 7053: + return VUID_WRAP(VUID-PrimitiveTriangleIndicesEXT-PrimitiveTriangleIndicesEXT-07053); + case 7055: + return VUID_WRAP(VUID-PrimitiveTriangleIndicesEXT-PrimitiveTriangleIndicesEXT-07055); + case 7056: + return VUID_WRAP(VUID-PrimitiveTriangleIndicesEXT-PrimitiveTriangleIndicesEXT-07056); case 7102: return VUID_WRAP(VUID-StandaloneSpirv-MeshEXT-07102); case 7320: